Business operating system for Claude Code — 57 skills, 21 agents, smart daemon. Unified inbox (WhatsApp/Email/Slack/Telegram), autonomous PR merge, full-AWS monitoring, revenue (Stripe+RevenueCat), e-commerce (Shopify), marketing (Klaviyo/Meta/GA4), voice (Bland/ElevenLabs), APM (Datadog/NewRelic/OTEL), YOLO mode.
Shell agent-teamsai-agentai-automationanthropicautonomous-agentawsbusiness-operationsclaudeclaude-codeclaude-code-plugindevopsecommerceinbox-zeroklaviyomarketingmcpmorning-briefingpr-automationshopifyterminal
chore: migrate ops wording to cliproxy and add reauth 2fa
mission-worker7cd9998dd7c1
diff --git a/claude-ops/CHANGELOG.md b/claude-ops/CHANGELOG.mdindex 5d8d703e..fd50cecf 100644--- a/claude-ops/CHANGELOG.md+++ b/claude-ops/CHANGELOG.md@@ -15,9 +15,9 @@### Changed- **ops-accounts multi-provider plan:** every provider (Claude, Grok, OpenAI/Codex, Factory, Cursor) gets Anthropic-shaped OAuth, reauth, util, switch/LB under `/ops:accounts`.-- **CRS cherry-pick path:** absorb gateway + Claude seat policy + Grok OAuth proxy into ops-accounts so default installs need no claude-relay-service; external CRS stays advanced-only. See `docs/ops/OPS-ACCOUNTS-VISION.md`.-- **ops-accounts seat policy (local):** `seat-policy-tick.mjs` applies conservative 5h/7d schedulable thresholds to local seat-state without CRS; `ops-accounts seats tick`.-- **ops-accounts local seat-state:** `seat-state.mjs` file-backed multi-provider seat store (schedulable/util) for no-CRS policy backend; `ops-accounts seats` command.+- **CLIProxyAPI gateway path:** absorb gateway + Claude seat policy + Grok OAuth proxy into ops-accounts so default installs use CLIProxyAPI-compatible relay behavior without requiring the deprecated relay stack; external legacy CRS stays advanced-only. See `docs/ops/OPS-ACCOUNTS-VISION.md`.+- **ops-accounts seat policy (local):** `seat-policy-tick.mjs` applies conservative 5h/7d schedulable thresholds to local seat-state without a relay; `ops-accounts seats tick`.+- **ops-accounts local seat-state:** `seat-state.mjs` file-backed multi-provider seat store (schedulable/util) for local policy backend; `ops-accounts seats` command.- **ops-accounts skill merge:** `/ops:accounts` is canonical multi-provider entry; `/ops:rotate` and `/ops:rotate-setup` are aliases. Expanded `bin/ops-accounts` (status/util/switch/refresh/reauth/setup/crs). Added `grok-reauth-egress.sh` residential cascade (EFG SOCKS → Bright Data tiers) for Grok device OAuth.- **Required companion co-install:** companions essential to ops commands are no longer optional setup prompts. `plugin-dependencies.json` marks `desktop-act`, `gsd`, `gstack` (skills clone), `superpowers`, and `feature-dev` as `required: true` with `updateWithOps: always` and `essentialFor` command lists. `scripts/install-companions.sh` always installs missing required companions (even under `--update-only`), and `--status` exits 1 when any required companion is missing. `/ops:setup` Step 2b runs the SSOT script without Skip for required deps. `/ops:update` step 9 co-installs the full required set (still skippable only via `--no-companions` / `OPS_SKIP_COMPANIONS=1`). Adds `scripts/install-gstack-companion.sh` (env-templated `GSTACK_REPO`/`GSTACK_DIR`/`GSTACK_BRANCH`).diff --git a/claude-ops/README.md b/claude-ops/README.mdindex 2a303bad..79cd8c06 100644--- a/claude-ops/README.md+++ b/claude-ops/README.md@@ -60,8 +60,8 @@ Per-repo budget caps (default 3/hour), single-flight locks, content-hash dedup,| ------------------- | --------------------------------------------------------------------------------- || `/ops` | Interactive command center dashboard (visual HQ) || `/ops:dash` | Same as `/ops` — pixel-art dashboard with hotkey navigation |-| `ops-healify-dash` | Dedicated Healify.ai KPI, growth, App Store, reliability, infra, repo, agent, MCP, and plugin command center |-| `ops-healify-bi-refresh` | Warms the Healify BI cache used by `ops-healify-dash` from EAS, ASC, AppsFlyer, Amplitude, Sentry, BetterStack, Linear, QA, and ops-dashboard sources |+| `ops-client-dash` | Dedicated client KPI, growth, App Store, reliability, infra, repo, agent, MCP, and plugin command center |+| `ops-client-bi-refresh` | Warms the client BI cache used by `ops-client-dash` from EAS, ASC, AppsFlyer, Amplitude, Sentry, BetterStack, Linear, QA, and ops-dashboard sources || `/ops:setup` | Interactive setup wizard — installs CLIs, configures channels, builds registry || `/ops:go` | Morning briefing — all systems in one dashboard || `/ops:next` | Priority-ordered next action (fires > comms > PRs > sprint > GSD) |diff --git a/claude-ops/bin/ops-accounts b/claude-ops/bin/ops-accountsindex 1f8efee0..d02b45cf 100755--- a/claude-ops/bin/ops-accounts+++ b/claude-ops/bin/ops-accounts@@ -91,10 +91,10 @@ PYelseecho "grok-cli-auth-proxy not configured or unreachable (set GROK_PROXY_URL)"fi- if [[ -n "$CRS_GROK_BASE_URL" ]] && curl -fsS -m 2 -o /dev/null -w "crs_grok_hop HTTP %{http_code}\n" "${CRS_GROK_BASE_URL%/}/models" 2>/dev/null; then+ if [[ -n "$CRS_GROK_BASE_URL" ]] && curl -fsS -m 2 -o /dev/null -w "cliproxy_grok_hop (legacy CRS_GROK_BASE_URL) HTTP %{http_code}\n" "${CRS_GROK_BASE_URL%/}/models" 2>/dev/null; then:else- echo "crs_grok_hop: not reachable (optional thin hop)"+ echo "cliproxy_grok_hop: not reachable (optional thin hop; legacy env CRS_GROK_BASE_URL)"fi}# --- CLIProxyAPI backend (optional local multi-provider proxy) ---@@ -299,14 +299,14 @@ PYif [[ -x "$ROT/crs-priority-daemon.sh" ]]; thenbash "$ROT/crs-priority-daemon.sh" --status 2>&1 | head -40else- echo "CRS priority script missing (optional)"+ echo "cliproxy / CLIProxyAPI priority script missing (optional; legacy filename retained)"fi;;crs-tick)if [[ -x "$ROT/crs-priority-daemon.sh" ]]; thenbash "$ROT/crs-priority-daemon.sh" "$@"else- echo "CRS priority script missing"; exit 1+ echo "cliproxy / CLIProxyAPI priority script missing (legacy filename retained)"; exit 1fi;;seats)diff --git a/claude-ops/bin/ops-bg b/claude-ops/bin/ops-bgindex 59ef2675..7592bd5a 100755--- a/claude-ops/bin/ops-bg+++ b/claude-ops/bin/ops-bg@@ -86,9 +86,9 @@ cmd_dispatch() {exit 2filocal prompt="$*"- # CRS routing: every bg session ops-bg spawns inherits the sanctioned overlay+ # cliproxy / CLIProxyAPI routing: every bg session ops-bg spawns inherits the sanctioned overlay# (ANTHROPIC_BASE_URL + CLAUDE_CODE_OAUTH_TOKEN). Preflight refuses to spawn if- # the overlay is half-set or the key is 401-rejected. Opt out with CRS_OVERLAY="".+ # the overlay is half-set or the key is 401-rejected. Opt out with CRS_OVERLAY="" (legacy env name).local CRS_OVERLAY="${CRS_OVERLAY-$HOME/.claude/crs-session-settings.json}"local -a crs_args=()if [ -n "$CRS_OVERLAY" ]; thendiff --git a/claude-ops/bin/ops-fleet b/claude-ops/bin/ops-fleetindex ce1f6bd1..d21357ee 100755--- a/claude-ops/bin/ops-fleet+++ b/claude-ops/bin/ops-fleet@@ -1,14 +1,14 @@#!/usr/bin/env bash# fleet-status.sh — polished fleet dashboard.# Unifies everything `claude agents --json`, `claude daemon status`, the daemon roster.json,-# each session's job state.json, the rotation daemon state, and CRS expose — per session:-# token type (oauth/api/bedrock/crs) · CRS routing · agent type · effort/model · respawn-attempts+# each session's job state.json, the rotation daemon state, and CLIProxyAPI-compatible relay expose — per session:+# token type (oauth/api/bedrock/cliproxy) · relay routing · agent type · effort/model · respawn-attempts# · repo · uptime · last-update · last-write · leased account · live detail · last error.# Read-only. No mutation. Usage: fleet-status.sh [--all] [--watch [secs]] [--no-color]set -uo pipefailCLAUDE_BIN="${CLAUDE_BIN:-/usr/local/bin/claude}"-CRS_PORT="${CRS_PORT:-3005}" # CRS (Claude Relay Service) listen port (3000 = legacy)+CRS_PORT="${CRS_PORT:-3005}" # CLIProxyAPI-compatible relay listen port (legacy env; 3000 = legacy)AR="$HOME/.claude/scripts/account-rotation"PROJ="$HOME/.claude/projects/-Users-<USER>"EC2_CACHE="${FLEET_EC2_CACHE:-$HOME/.claude/state/fleet-ec2.json}"@@ -245,7 +245,7 @@ title=" FLEET DASHBOARD "; clock=f" {time.strftime('%Y-%m-%d %H:%M:%S')} "fill=W-len(title)-len(clock)print(f"{B}╔{title}{'═'*fill}{clock}╗{R}")print(f" {B}DAEMON{R} pid {d_pid} · v{d_ver} · up {d_up} · {origin} · sock {sock} · workers {d_work} · roster {d_roster}")-print(f" {B}CRS{R} /health {hb} · allowlist {crs_allow_label} · 6m: done~{crs_done} {YEL}429:{crs_429}{R} {RED}529:{crs_529}{R} · keychain-swap {rh}")+print(f" {B}CLIPROXY{R} /health {hb} · legacy allowlist {crs_allow_label} · 6m: done~{crs_done} {YEL}429:{crs_429}{R} {RED}529:{crs_529}{R} · keychain-swap {rh}")if ec2_on:if ec2_ok:m=re.search(r'(\d+) running.*?(\d+) in roster',ec2_daemon)@@ -255,7 +255,7 @@ if ec2_on:print(f" {B}EC2 FRA{R} {GRN}✓{R} {ec2_n} sess · {dw}cache {agestr} ago{stale} {D}· {ec2_inst}{R}")else:print(f" {B}EC2 FRA{R} {RED}✗ {ec2_err or 'no cache'}{R} {D}(refreshing in background…){R}")-print(f" {D}token {GRN}oauth{D}=Max {YEL}api{D}=metered {MAG}bedrock{D}=AWS {CYN}crs{D}=relay→Max pool crs: {CYN}▶{D}routing ●listed ·direct {BLU}fra{D}=EC2 session{R}")+print(f" {D}token {GRN}oauth{D}=Max {YEL}api{D}=metered {MAG}bedrock{D}=AWS {CYN}cliproxy{D}=relay→Max pool relay: {CYN}▶{D}routing ●listed ·direct {BLU}fra{D}=EC2 session{R}")print(f"{B}╚{'═'*(W)}╝{R}")# ---- group by state ----@@ -335,9 +335,9 @@ for email,info in sorted(accts.items(),key=lambda x:-(x[1].get("lastUtilization"bill=(info.get("lastBilling",{}) or {}).get("billing_type","—"); extra=" +x" if (info.get("lastBilling",{}) or {}).get("extra_usage_enabled") else ""print(f" {email:34} {barv} {col}{str(pct if pct is not None else '—'):>3}%{R} {D}{reset_in(u):>9} · act {iso_ago(info.get('lastActive')):>6} · {bill}{extra}{R}")-# ---- CRS proxy routing ----+# ---- CLIProxyAPI/legacy CRS proxy routing ----if acct_hits:- print(f"{B}CRS ROUTING{R} {D}(accounts used last 6m){R}")+ print(f"{B}CLIPROXY ROUTING{R} {D}(accounts used last 6m){R}")line=" "+" ".join(f"{CYN}{a}{R}{D}:{n}{R}" for a,n in sorted(acct_hits.items(),key=lambda x:-x[1]))print(line)PYdiff --git a/claude-ops/bin/ops-no-rm-rf-anchor b/claude-ops/bin/ops-no-rm-rf-anchorindex 193da2c9..58b9cebf 100755--- a/claude-ops/bin/ops-no-rm-rf-anchor+++ b/claude-ops/bin/ops-no-rm-rf-anchor@@ -84,23 +84,21 @@ DANGER=$(set -f; while IFS= read -r seg; do# Tokenize on spaces (best-effort — won't perfectly handle quoted-with-spaces but covers the patterns specified).for tok in $rest; do# Skip flags.- case "$tok" in- -*) continue ;;- esac+ if [[ "$tok" == -* ]]; then+ continue+ fi# Strip surrounding quotes for comparison.bare="${tok%\"}"; bare="${bare#\"}"bare="${bare%\'}"; bare="${bare#\'}"- case "$bare" in- "" | "/" | "/*" | "/." | \- "~" | "~/" | "~/*" | \- '$HOME' | '$HOME/' | '$HOME/*' | \- ".." | "../" | "../*" | \- ".")- DANGER="$bare"- break- ;;- esac+ if [[ "$bare" == "" || "$bare" == "/" || "$bare" == "/*" || "$bare" == "/." || \+ "$bare" == "~" || "$bare" == "~/" || "$bare" == "~/*" || \+ "$bare" == '$HOME' || "$bare" == '$HOME/' || "$bare" == '$HOME/*' || \+ "$bare" == ".." || "$bare" == "../" || "$bare" == "../*" || \+ "$bare" == "." ]]; then+ DANGER="$bare"+ break+ fi# Also catch quoted $HOME variants where the shell hasn't expanded them.if [[ "$tok" == '"$HOME"' || "$tok" == '"$HOME"/' || "$tok" == '"$HOME"/*' ]]; thenDANGER="$tok"; breakdiff --git a/claude-ops/docs/healify-dashboard.md b/claude-ops/docs/healify-dashboard.mdindex 7e3f633c..b547c957 100644--- a/claude-ops/docs/healify-dashboard.md+++ b/claude-ops/docs/healify-dashboard.md@@ -1,69 +1,29 @@-# Healify Dashboard+# Client Dashboard-`ops-healify-dash` is a dedicated Healify.ai command center for terminal and tmux workflows. It is read-only by default and designed for dark terminals with an orange Healify.ai brand treatment.+This repository is public, so client-specific dashboard details live outside the committed docs. Keep this page generic and avoid real customer, operator, or brand identifiers.-## Commands+## Public Contract-```bash-ops-healify-dash --once-ops-healify-dash --watch-ops-healify-dash --refresh-ops-healify-dash --json-ops-healify-bi-refresh-```--`--refresh` refreshes bounded local caches first. `--watch` rerenders on an interval. Neither mode changes tmux focus or selects another window.--`ops-dash` can explicitly delegate to this surface:--```bash-OPS_HEALIFY_DASH=1 ops-dash-```+A client dashboard may aggregate only local, already-configured sources and should remain read-only by default. Watch modes may rerender on an interval, but they must not change tmux focus or select another window.## Data Sources-The dashboard aggregates only local, already-configured sources:+Allowed source categories for public documentation:-- `ops:projects` cache: `$OPS_DATA_DIR/cache/projects_health.json`-- statusline KPI cache: `~/.claude/state/statusline-kpi.json`-- live endpoint health cache: `~/.cache/agent-hub/infra-health.json`-- AWS ECS health cache: `~/.cache/agent-hub/aws-ecs-health.json`-- plugin inventory: `~/.cache/agent-hub/plugin-inventory.json`-- Healify BI cache: `~/.cache/agent-hub/healify-bi.json`-- runtime health: `~/.cache/agent-hub/fleet/runtime-health.txt`-- local Healify git repos under `~/Projects/healify-workspace`+- project health caches+- statusline KPI caches+- live endpoint health caches+- cloud service health caches+- plugin inventory+- runtime health summaries- important ops and agent-hub logs-`ops-healify-bi-refresh` warms the BI cache from authenticated local and ops-dashboard sources when available:--- EAS latest iOS builds-- App Store Connect app versions, build numbers, processing state, and reviews-- App Store Connect sales/download status, clearly marked blocked when the vendor number is missing-- AppsFlyer installs, organic split, loyal users, and cost-- Amplitude active/new-user status-- Sentry unresolved issue counts by Healify project-- BetterStack uptime/incidents-- Linear priority issue counts-- mobile QA and Pact/contract health from the operating dashboard integration source--It shows business KPIs, acquisition and marketing state, App Store/TestFlight release state, reviews, project execution state, service health, ECS deployment status, repo drift, agent/MCP/CLI readiness, plugin coverage, and recent important logs.+Authenticated business intelligence sources, app-store metadata, analytics providers, incident providers, and issue trackers must be documented in private runbooks instead of this public repo.## Branding-The 1:1 pixel icon lives at:--```text-assets/healify-pixel-logo.svg-```--It follows the current Healify.ai app-icon direction: orange/pink rounded square, high contrast, and a white heart/pill health mark. The terminal dashboard renders the same concept as ANSI pixel blocks.+Do not commit client-specific names, logos, colors, or brand assets in this public documentation. Use private docs for client-specific dashboard branding.## Tmux Integration-Any tmux binding should open this dashboard only through explicit user action, for example:--```tmux-bind v display-popup -E -w 96% -h 92% "ops-healify-dash --watch"-```--Background watchdogs, prewarmers, doctors, and repair hooks must not call `select-window` or `switch-client`. While a user is working in a window, focus should remain there unless the user explicitly presses a navigation key.+Any tmux binding should open a client dashboard only through explicit user action. Background watchdogs, prewarmers, doctors, and repair hooks must not call `select-window` or `switch-client`. While a user is working in a window, focus should remain there unless the user explicitly presses a navigation key.diff --git a/claude-ops/docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.md b/claude-ops/docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.mdindex c2572d11..aae5e5da 100644--- a/claude-ops/docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.md+++ b/claude-ops/docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.md@@ -11,12 +11,12 @@ Public-repo clean: no host emails, IPs, or personal account lists.|--------|------------------------|------------------------|| `rotate.mjs` | ~8981 lines | ~5275 lines (after this pass: soft captcha hooks) || Captcha helper stack | Present (full) | **Ported this pass** |-| CRS reconcilers / priority | Present (ops + host forks) | Present (portable subset) |+| CLIProxyAPI/legacy CRS reconcilers / priority | Present (ops + host forks) | Present (portable subset) || Magic-link autoloop + reauth env | Host thicker; plugin has portable `reauth-env.mjs` | Present (PR #723 era) || Production systemd/user units | Still point at **host** path | Templates only (`%h` / `$CLAUDE_PLUGIN_ROOT`) |-**CRS is optional.** Standalone rotate-magic (keychain + `rotate.mjs` /-`rotate-magic.mjs`) is enough for one or a few accounts. CRS is multi-account+**CLIProxyAPI/legacy relay is optional.** Standalone rotate-magic (keychain + `rotate.mjs` /+`rotate-magic.mjs`) is enough for one or a few accounts. CLIProxyAPI-compatible relay mode is multi-accountload balancing / rate-limit spreading for a relay pool.## Line-count delta (key modules)@@ -96,7 +96,7 @@ After this pass, cascade **modules** exist; full host `rotate.mjs` still has:6. Residential Chrome / PAC / EFG proxy args for reauth browser7. `secrets-bootstrap` before solver use8. `virtual-display` ensure for headed seats-9. CRS push helpers (`--sync-crs-all` style)+9. Legacy CRS push helpers (`--sync-crs-all` compatibility style)10. Stricter OAuth callback / credential race handling## Minimal port set (standalone plugin rotate-magic)@@ -115,7 +115,7 @@ After this pass, cascade **modules** exist; full host `rotate.mjs` still has:- (keeps after-magic-link verify in `finishMagicLinkLogin`)- `rotate-magic.mjs` runs ensure before spawning `rotate.mjs`- Thin `rotate-magic.mjs` entry → `rotate.mjs --magic-link`-- CRS optional UX in `/ops:rotate-setup` and `/ops:rotate` skills+- CLIProxyAPI/legacy relay optional UX in `/ops:rotate-setup` and `/ops:rotate` skills- Units: `__tests__/captcha-cascade.test.mjs`, `__tests__/ensure-rotate-captcha-hooks.test.mjs`**Apply hooks on a checkout**diff --git a/claude-ops/docs/ops/OPS-ACCOUNTS-GATEWAY.md b/claude-ops/docs/ops/OPS-ACCOUNTS-GATEWAY.mdindex 3644a4dc..433260d4 100644--- a/claude-ops/docs/ops/OPS-ACCOUNTS-GATEWAY.md+++ b/claude-ops/docs/ops/OPS-ACCOUNTS-GATEWAY.md@@ -2,9 +2,9 @@**Status:** skeleton shipped (`scripts/account-rotation/ops-accounts-gateway.mjs`).-**Goal:** Optional thin OpenAI-compat gateway so fleets that today set-`base_url=http://127.0.0.1:3005` (CRS) can point at a plugin-owned process-**without** installing weishaw/claude-relay-service.+**Goal:** Optional thin OpenAI-compat CLIProxyAPI gateway so fleets that today set+`base_url=http://127.0.0.1:3005` for cliproxy / CLIProxyAPI can point at a plugin-owned process+**without** installing any deprecated relay stack.## What it does (skeleton)@@ -43,7 +43,7 @@ CRS_BASE_URL → OPS_ACCOUNTS_GATEWAY_URL# harnesses keep OpenAI-compat client shape```-External CRS remains **advanced-only** for operators who want the full CRS admin UI.+Deprecated relay stacks remain **advanced-only** for operators who want the old admin UI.## Build order@@ -55,5 +55,5 @@ External CRS remains **advanced-only** for operators who want the full CRS admin## License-If any line is copied from CRS (MIT), keep attribution. Prefer rewrite of the-proxy surface over vendoring the monorepo.+If any line is copied from a deprecated MIT relay, keep attribution. Prefer+rewriting the proxy surface over vendoring the monorepo.diff --git a/claude-ops/docs/ops/OPS-ACCOUNTS-VISION.md b/claude-ops/docs/ops/OPS-ACCOUNTS-VISION.mdindex 64292cd7..54e3df3f 100644--- a/claude-ops/docs/ops/OPS-ACCOUNTS-VISION.md+++ b/claude-ops/docs/ops/OPS-ACCOUNTS-VISION.md@@ -21,7 +21,7 @@ Ideal box: **public claude-ops plugin only** — no host-only `grok-rotate`, `co## Anthropic reference stack (what “done” looks like)-Claude already has (in plugin + optional CRS):+Claude already has (in plugin + optional cliproxy / CLIProxyAPI relay):| Layer | Role ||-------|------|@@ -31,7 +31,7 @@ Claude already has (in plugin + optional CRS):| Reauth | magic-link-autoloop + captcha cascade (residential egress) || Util | 5h / 7d utilization queries || Switch | keychain swap / force-rotate |-| Optional LB | CRS pool priority, 429 cooldown, token feed |+| Optional LB | cliproxy / CLIProxyAPI pool priority, 429 cooldown, token feed |**Every other provider gets the same layers**, with provider-native OAuth and util APIs.@@ -39,21 +39,21 @@ Claude already has (in plugin + optional CRS):| Provider | OAuth / capture | Reauth | Util | Switch / LB notes ||----------|-----------------|--------|------|-------------------|-| **Claude (Anthropic)** | magic-link + captcha | magic-link-autoloop | 5h/7d | keychain daemon; CRS optional |-| **Grok (xAI SuperGrok)** | device-code + Google (dcli) | residential cascade (EFG SOCKS → Bright Data residential → ISP → mobile) | weekly / 429 | `grok-cli-auth-proxy` RR; CRS is thin relay only |+| **Claude (Anthropic)** | magic-link + captcha | magic-link-autoloop | 5h/7d | keychain daemon; cliproxy / CLIProxyAPI relay optional |+| **Grok (xAI SuperGrok)** | device-code + Google (dcli) | residential cascade (EFG SOCKS → Bright Data residential → ISP → mobile) | weekly / 429 | `grok-cli-auth-proxy` RR; optional CLIProxyAPI-compatible hop only || **OpenAI / Codex** | OAuth / API key | codex OAuth bridge | usage API | `codex-rotate` absorbed || **CLIProxyAPI (local multi-provider pool)** | per-provider OAuth login flows (browser/device-code) | proxy-internal token refresh | management API (pool-level) | single static binary + flat-file auth-dir; chosen over sub2api (no Postgres/Redis to maintain). Optional local backend: `providers/cliproxy.mjs` adapter reports auth-dir inventory; `bin/ops-accounts` prints per-provider counts + `util <provider>` views. Env overrides: `CLIPROXYAPI_HOME`, `CLIPROXYAPI_AUTH_DIR`, `CLIPROXYAPI_BASE_URL`. || **Factory** | provider OAuth / tokens | native reauth | quota-feed-factory patterns | full adapter || **Cursor** | Cursor account OAuth | browser/device OAuth | plan limits if available | full adapter || **Extensible** | adapter interface | same contract | best-effort | `provider-env` / `provider-router` |-### Grok + CRS (do not mis-sell)+### Grok + CLIProxyAPI-compatible relay (do not mis-sell)-Grok Build can use `CRS_GROK_BASE_URL` for its CRS-compatible endpoint. CRS **does not** own a SuperGrok account pool. It forwards to the configured **`grok-cli-auth-proxy`** (`GROK_PROXY_URL`), which holds OAuth seats and round-robins. ops-accounts status must show:+Grok Build can use `CRS_GROK_BASE_URL` as a legacy env alias for its CLIProxyAPI-compatible endpoint. The optional relay hop **does not** own a SuperGrok account pool. It forwards to the configured **`grok-cli-auth-proxy`** (`GROK_PROXY_URL`), which holds OAuth seats and round-robins. ops-accounts status must show:-1. CRS hop present/absent-2. Proxy RR seat health + exhaust cooldowns-3. Slot ↔ `auth.json` sync+1. CLIProxyAPI-compatible hop present/absent+2. Proxy RR seat health + exhaust cooldowns+3. Slot ↔ `auth.json` sync## Adapter contract (every provider)@@ -75,7 +75,7 @@ No secrets in skill output. Provider-scoped vault service names. Env-templated p|-------|------|----------------|| **0** | Skill merge + contract — `/ops:accounts` owns verbs; rotate/setup are aliases; thin router over existing engines | One entrypoint; multi-provider status || **1** | Claude plugin parity (captcha cascade, standalone reauth) | Host not required for Claude magic-link |-| **2** | Claude CRS optional | Detect; never required for single seat |+| **2** | Claude CLIProxyAPI/legacy relay optional | Detect; never required for single seat || **3** | Grok adapter complete (plugin reauth + residential cascade + slot sync + proxy status) | Multi-seat healthy under weekly cap without host-only ops || **4** | OpenAI/Codex + Factory adapters | same verbs || **5** | Cursor adapter | same verbs |@@ -85,51 +85,51 @@ No secrets in skill output. Provider-scoped vault service names. Env-templated p-## CRS cherry-pick (so people do not need CRS)+## cliproxy / CLIProxyAPI gateway path (legacy aliases retained)-### Split CRS into three jobs+### Split cliproxy-compatible behavior into three jobs1. **API gateway** — OpenAI-compat `:3005`, `cr_` keys, harness routing2. **Claude seat LB** — schedulable pool, 429/weekly park, token feed-3. **Grok thin hop** — forward to host SuperGrok OAuth proxy (not a CRS pool)+3. **Grok thin hop** — forward to host SuperGrok OAuth proxy (not a Claude relay pool)ops-accounts absorbs (1)+(2) as optional **ops-accounts-gateway** + **policy daemons**,and treats (3) as the **Grok provider backend** (port `grok-cli-auth-proxy` into-the plugin). External CRS remains advanced-only.+the plugin). External legacy relay stacks remain advanced-only compatibility paths.-### Already portable in this repo (rename, do not re-clone CRS)+### Already portable in this repo (legacy filenames retained for compatibility)- `crs-priority-daemon.mjs` — pool priority / schedulable policy- `crs-429-cooldown.mjs` — rate-limit cooldown- `crs-401-refresher.mjs` — proactive refresh- `crs-token-feed.mjs` — vault → pool-- `crs-pool-config.mjs`, `crs-health-watch.mjs`, `crs-bedrock-guard.mjs` — optional+- `crs-pool-config.mjs`, `crs-health-watch.mjs`, `crs-bedrock-guard.mjs` — optional-These currently *call* CRS admin APIs. Dual-mode target:+These legacy-named modules currently call cliproxy-compatible admin APIs. Dual-mode target:-- `backend=crs` — talk to external CRS (compat)-- `backend=local` — file/SQLite seat state, no Docker+- `backend=crs` — legacy alias for the cliproxy-compatible admin API backend+- `backend=local` — file/SQLite seat state, no Docker### Build sequence| Step | Deliverable ||------|-------------|-| A | Local seat-state backend for policy daemons (no CRS) |+| A | Local seat-state backend for policy daemons (no relay required) || B | Plugin-owned Grok OAuth proxy (today’s `:31845` logic) || C | Optional OpenAI-compat gateway replacing `:3005` for most harnesses |-| D | Docs: migrate `CRS_BASE_URL` → gateway; CRS optional advanced |+| D | Docs: migrate legacy `CRS_BASE_URL` aliases → gateway; cliproxy optional advanced |### License-CRS lineage on this box is **MIT**. Cherry-pick with attribution; prefer-reimplemented policy + small gateway over vendoring the full CRS monorepo-(Redis/Grafana/admin SPA out of default install).+Historical relay lineage on this box is **MIT**. Cherry-pick with attribution;+prefer reimplemented policy + small gateway over vendoring deprecated relay+monorepos (Redis/Grafana/admin SPA out of default install).### Non-goals-- Shipping Redis/Prometheus/Grafana as required deps-- Multi-tenant CRS admin SPA-- Claiming Grok seats live inside a Claude-style CRS account table+- Shipping Redis/Prometheus/Grafana as required deps+- Multi-tenant relay admin SPA+- Claiming Grok seats live inside a Claude-style relay account table## Risks@@ -143,13 +143,13 @@ reimplemented policy + small gateway over vendoring the full CRS monorepo## Non-goals (immediate)-- Full CRS fork for every vendor-- Deleting host trees before parity proof-- Hard-requiring CRS+- Full deprecated-relay fork for every vendor+- Deleting host trees before parity proof+- Hard-requiring a relay## Related-- `docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.md`-- `scripts/account-rotation/CAPTCHA-CASCADE.md`-- Skills: `ops-accounts` (canonical), `ops-rotate` / `ops-rotate-setup` (aliases)-- Local plan: `Projects/memory/plans/2026-07-30T1557Z-public-ops-plugin-only.md`+- `docs/ops/HOST-VS-PLUGIN-ROTATE-GAP.md`+- `scripts/account-rotation/CAPTCHA-CASCADE.md`+- Skills: `ops-accounts` (canonical), `ops-rotate` / `ops-rotate-setup` (aliases)+- Local plan: `Projects/memory/plans/2026-07-30T1557Z-public-ops-plugin-only.md`diff --git a/claude-ops/docs/pocket-ec2-deployment.md b/claude-ops/docs/pocket-ec2-deployment.mdindex 6bf131e9..47bee277 100644--- a/claude-ops/docs/pocket-ec2-deployment.md+++ b/claude-ops/docs/pocket-ec2-deployment.md@@ -1,18 +1,16 @@-# Pocket pipeline — EC2 deployment on dev-sandbox+# Pocket pipeline — EC2 deployment-Pocket runs entirely on the dev-sandbox EC2 instance (i-0b207951a7dff925d,-m8g.2xlarge). This is an approved exception to the no-EC2 rule; pocket-workloads are too persistent for a laptop.+Pocket can run on a dedicated EC2 instance. This is an approved exception to+the no-EC2 rule; pocket workloads are too persistent for a laptop.## Access-Web UI: `https://dev-sandbox.tail6aeed8.ts.net` (Tailscale only — never-exposed via public DNS or Cloudflare). Auth is enforced by the+Web UI: expose the local service over a private Tailscale HTTPS hostname only —+never via public DNS or Cloudflare. Auth is enforced by the`Tailscale-User-Login` header injected by `tailscale serve`; onlythe configured owner identity is admitted (set via `TAILSCALE_USER` env in the systemd unit).-SSH: `ssh dev-ts` (ec2-user, pem at `~/.ssh/dev-sandbox-2026-05-17.pem`,-Tailscale IP 100.109.217.31).+SSH: use the operator's private host alias and key outside this public repo.## Systemd units@@ -34,9 +32,9 @@ plugin upgrade: `bash claude-ops/scripts/systemd/install-systemd-units.sh`.## Tailscale serve`tailscale serve --https=443 http://127.0.0.1:7777` is configured on-dev-sandbox (runs as root via `sudo tailscale serve`). It terminates TLS,-injects the identity header, and proxies to gunicorn. Config persists across-reboots. Verify with `sudo tailscale serve status`.+the private EC2 host (runs as root via `sudo tailscale serve`). It terminates+TLS, injects the identity header, and proxies to gunicorn. Config persists+across reboots. Verify with `sudo tailscale serve status`.## State directorydiff --git a/claude-ops/docs/safety-hooks.md b/claude-ops/docs/safety-hooks.mdindex a9840418..012e4b76 100644--- a/claude-ops/docs/safety-hooks.md+++ b/claude-ops/docs/safety-hooks.md@@ -119,7 +119,7 @@ bash claude-ops/tests/test-safety-hooks.sh## bedrock-billing-guard (PreToolUse) — metered-spend protection`hooks/bedrock-billing-guard.mjs` hard-stops any Claude session that is **measured-to be billing AWS Bedrock** (metered per-token spend, not the Max/CRS account+to be billing AWS Bedrock** (metered per-token spend, not the Max/CLIProxyAPI accountpool). It is the agent-facing complement to the rotation daemon's`scripts/account-rotation/bedrock-watchdog.mjs`, which detects + force-swaps suchsessions off Bedrock within ~45s.@@ -139,8 +139,8 @@ network/env measurement.| `warn` | Never block — inject a one-line non-blocking warning into the agent's context on each tool call. || `off` | No-op. For users who intentionally run **Bedrock unrestricted**. |-When OAuth headroom exists, the watchdog respawns the session onto CRS/OAuth and-clears the flag, so the block lifts on its own. The guard is **fail-open**: any+When OAuth headroom exists, the watchdog respawns the session onto CLIProxyAPI/OAuth+(or a legacy CRS-compatible relay) and clears the flag, so the block lifts on its own. The guard is **fail-open**: anyerror, missing input, or unexpected state allows the tool (a money guard mustnever wedge an innocent session).diff --git a/claude-ops/hooks/bedrock-fallback-guard.mjs b/claude-ops/hooks/bedrock-fallback-guard.mjsindex 5daabbf1..3c607307 100644--- a/claude-ops/hooks/bedrock-fallback-guard.mjs+++ b/claude-ops/hooks/bedrock-fallback-guard.mjs@@ -18,19 +18,19 @@ try {if (settings.mixedProvider) {block(- 'Claude routing is unsafe: settings.json has both Bedrock and CRS/OAuth enabled. Run `claude-stack route --mode crs-oauth` or explicitly confirm Bedrock with `claude-stack route --mode bedrock-confirmed --confirm-metered-bedrock --reason "<reason>"`.',+ 'Claude routing is unsafe: settings.json has both Bedrock and CLIProxyAPI/legacy relay OAuth enabled. Run `claude-stack route --mode crs-oauth` (legacy mode name) or explicitly confirm Bedrock with `claude-stack route --mode bedrock-confirmed --confirm-metered-bedrock --reason "<reason>"`.',);}if (settings.bedrock && (route.mode !== 'bedrock-confirmed' || !status.bedrockConfirmationActive)) {block(- 'Bedrock fallback blocked. Bedrock is metered AWS usage, and OAuth/CRS is unavailable or not selected. Confirm only for this session with `claude-stack route --mode bedrock-confirmed --reason "<why OAuth is unavailable>" --ttl-minutes 60 --confirm-metered-bedrock`, or restore OAuth with `claude-stack route --mode crs-oauth`.',+ 'Bedrock fallback blocked. Bedrock is metered AWS usage, and CLIProxyAPI/legacy relay OAuth is unavailable or not selected. Confirm only for this session with `claude-stack route --mode bedrock-confirmed --reason "<why OAuth is unavailable>" --ttl-minutes 60 --confirm-metered-bedrock`, or restore OAuth with `claude-stack route --mode crs-oauth` (legacy mode name).',);}if (route.mode === 'fail-closed') {block(- `Claude routing is fail-closed: ${route.reason || 'CRS OAuth is unavailable and Bedrock has not been explicitly confirmed'}. Run \`claude-stack doctor --json\` for current CRS/account status.`,+ `Claude routing is fail-closed: ${route.reason || 'CLIProxyAPI/legacy relay OAuth is unavailable and Bedrock has not been explicitly confirmed'}. Run \`claude-stack doctor --json\` for current relay/account status.`,);}diff --git a/claude-ops/scripts/account-rotation/CAPTCHA-CASCADE.md b/claude-ops/scripts/account-rotation/CAPTCHA-CASCADE.mdindex f8d2a025..be1a3846 100644--- a/claude-ops/scripts/account-rotation/CAPTCHA-CASCADE.md+++ b/claude-ops/scripts/account-rotation/CAPTCHA-CASCADE.md@@ -5,13 +5,13 @@ Contract for **unattended** re-auth owned by `magic-link-autoloop` →Agents must not spawn parallel `rotate.mjs` drivers against the same fleet(they thrash the global `.rotating` lock and the reauth browser profile).-**CRS is not required.** Captcha cascade runs in standalone rotate-magic.+**cliproxy / CLIProxyAPI relay is not required.** Captcha cascade runs in standalone rotate-magic.## Ownership| Component | Role ||-----------|------|-| `magic-link-autoloop` | Serial, one account per tick; opt-in (often CRS-flagged, but not required) |+| `magic-link-autoloop` | Serial, one account per tick; opt-in (often relay-flagged via legacy CRS config, but not required) || `rotate.mjs` / `rotate-magic.mjs` | Browser OAuth / magic-link / setup || `captcha-helper.mjs` | Token solvers + residential wait (`residualAfterWait`) || `visual-captcha-solver.mjs` | Vision tiles + desktop-act + VNC layers |diff --git a/claude-ops/scripts/account-rotation/README.md b/claude-ops/scripts/account-rotation/README.mdindex 6913f759..225cee5b 100644--- a/claude-ops/scripts/account-rotation/README.md+++ b/claude-ops/scripts/account-rotation/README.md@@ -54,7 +54,7 @@ Each account entry:| `dashlaneTokenPath` | no | If you store the token in Dashlane: `dl://<vault-name>/password`. || `extraUsageEnabled` | no | Set to `true` ONLY if the account has paid overage on. Triggers safety margin. || `capacityMultiplier` | no | Override per-account threshold (default 1.0 = standard Max 20x quota). |-| `crsAccountName` | no | CRS admin account name this vault entry maps to (for crs-token-feed / crs-priority). |+| `crsAccountName` | no | Legacy compatibility field for the cliproxy/CLIProxyAPI account name this vault entry maps to (for `crs-token-feed` / `crs-priority` compatibility). |## Signed staged Claude enrollment@@ -200,18 +200,18 @@ the coordination key. Trust files must remain outside all mutable auth roots.autoloop no longer dispatch authentication. They return a staged-enrollmenthandoff instead; no environment flag restores the old write path.-## CRS pool (optional)+## cliproxy / CLIProxyAPI pool (optional)-CRS = multi-account **load balancing / rate-limit spreading** via a relay pool.+cliproxy / CLIProxyAPI relay mode provides multi-account **load balancing / rate-limit spreading** via a relay pool.Useful mainly when many accounts share one API endpoint and you hit 429s.-Skip CRS for single/few-account keychain rotation.+Skip the relay for single/few-account keychain rotation.-If you run [claude-relay-service](https://github.com/anthropics/claude-relay-service) (CRS) alongside the rotator:+If you run a cliproxy relay alongside the rotator:-1. Add `crsAccountName` on each account (must match the account `name` in the CRS admin UI), or supply `crs.nameByVaultKey` in `config.json`.-2. Set `crs.enabled`, `crs.baseUrl`, and install the priority daemon: `scripts/install-crs-priority-agent.sh`.-3. On Linux/EC2, `crs-token-feed.mjs` propagates vault tokens into CRS (systemd timer when installed).-4. On macOS clients that reach a **remote** CRS via SSH, install the tunnel (requires `CRS_TUNNEL_SSH_HOST`):+1. Add `crsAccountName` on each account (legacy field name; must match the cliproxy account `name`), or supply `crs.nameByVaultKey` in `config.json`.+2. Set `crs.enabled`, `crs.baseUrl`, and install the priority daemon: `scripts/install-crs-priority-agent.sh` (legacy filename retained).+3. On Linux/EC2, `crs-token-feed.mjs` propagates vault tokens into the cliproxy pool (systemd timer when installed).+4. On macOS clients that reach a **remote** relay via SSH, install the tunnel (legacy env `CRS_TUNNEL_SSH_HOST` is still accepted):```bashCRS_TUNNEL_SSH_HOST=your-remote-host bash scripts/install-crs-fra-tunnel.sh@@ -219,8 +219,8 @@ If you run [claude-relay-service](https://github.com/anthropics/claude-relay-serPoint Claude Code at `http://127.0.0.1:3005/api` (or your `CRS_TUNNEL_LOCAL_PORT`).-See `config.example.json` → `crs` block for all tunables (`policy`, `fileVaultPath`, `containerName`, thresholds).-Configure via `/ops:rotate-setup` (detects CRS; never required).+See `config.example.json` → `crs` block for all tunables (`policy`, `fileVaultPath`, `containerName`, thresholds; legacy key names retained for compatibility).+Configure via `/ops:rotate-setup` (detects cliproxy / CLIProxyAPI; never required).## Keychain layout@@ -236,7 +236,7 @@ Override the keychain account name via `CLAUDE_ROTATOR_KEYCHAIN_ACCOUNT` if you| File | Purpose || --------------------------- | ------------------------------------------------------------------------- || `rotate.mjs` | Main rotation logic. CLI: `--status`, `--utilization`, `--to`, `--setup`. |-| `rotate-magic.mjs` | Thin entry → `rotate.mjs --magic-link` (standalone reauth, no CRS). |+| `rotate-magic.mjs` | Thin entry → `rotate.mjs --magic-link` (standalone reauth, no cliproxy required). || `captcha-cascade.mjs` | Post-verify captcha orchestration (token + visual + desktop-act + VNC). || `captcha-helper.mjs` | Pluggable token captcha solvers + residential wait. || `visual-captcha-solver.mjs` | Vision tile clicks, desktop-act, VNC layers. |diff --git a/claude-ops/scripts/account-rotation/claude-settings-mode.mjs b/claude-ops/scripts/account-rotation/claude-settings-mode.mjsindex 75e08691..4b520ad7 100644--- a/claude-ops/scripts/account-rotation/claude-settings-mode.mjs+++ b/claude-ops/scripts/account-rotation/claude-settings-mode.mjs@@ -15,14 +15,14 @@ function claudeSettingsPath() {return join(process.env.HOME || '', '.claude', 'settings.json');}-// ── CRS (Claude Relay Service) routing awareness ───────────────────────────────-// When a box routes through CRS, ANTHROPIC_BASE_URL points at the local relay+// ── CLIProxyAPI/legacy CRS-compatible relay routing awareness ──────────────────+// When a box routes through the relay, ANTHROPIC_BASE_URL points at the local relay// (e.g. http://127.0.0.1:3005/api) and CLAUDE_CODE_OAUTH_TOKEN holds a static// `cr_…` relay token. Both must travel together — a `cr_` token with no base URL// 401s. The Bedrock/OAuth settings mutators below were written for the raw-OAuth /-// Bedrock world and would silently strip ANTHROPIC_BASE_URL, stranding a CRS box.+// Bedrock world and would silently strip ANTHROPIC_BASE_URL, stranding a relay-routed box.// These helpers detect that routing and snapshot/restore the pair so a mode flip-// never breaks (or money-leaks past) a CRS-routed fleet.+// never breaks (or money-leaks past) a relay-routed fleet.function isCrsToken(t) {return typeof t === 'string' && t.startsWith('cr_');}diff --git a/claude-ops/scripts/account-rotation/config.example.json b/claude-ops/scripts/account-rotation/config.example.jsonindex 8c4ce6d7..21d27068 100644--- a/claude-ops/scripts/account-rotation/config.example.json+++ b/claude-ops/scripts/account-rotation/config.example.json@@ -30,7 +30,7 @@"refreshSession": false,"notifyOnRotation": true,"crs": {- "_comment": "Optional: claude-relay-service (CRS) pool integration. Off by default — set enabled=true and install via scripts/install-crs-priority-agent.sh. Map each rotator account to its CRS row with crsAccountName on the account entry (or crs.nameByVaultKey). Admin password is NOT stored here: set $CRS_ADMIN_PASSWORD or credential-store CRS-Admin-<adminUser>.",+ "_comment": "Optional: cliproxy / CLIProxyAPI pool integration. Off by default — set enabled=true and install via scripts/install-crs-priority-agent.sh (legacy filename). The crs* field names are retained as compatibility aliases. Map each rotator account to its relay row with crsAccountName on the account entry (or crs.nameByVaultKey). Admin password is NOT stored here: set $CRS_ADMIN_PASSWORD or credential-store CRS-Admin-<adminUser>.","enabled": false,"policy": "conservative","_policy_note": "conservative (default) deprioritizes on utilization/session warnings; max-out keeps accounts schedulable until genuine rate-limit or overload (util % is telemetry only). Override with $CRS_POLICY.",@@ -52,7 +52,7 @@"_thresholds_note": "Legacy conservative thresholds (off/on/floor) apply only when policy=conservative. Default max-out policy keeps accounts schedulable until genuine rate-limit or overload; utilization % is telemetry only. freshMinutes = max age of claudeUsage trusted for secondary signals.","logDir": null,"logTzOffset": "+00:00",- "_logDir_note": "Optional: crs-429-cooldown.mjs reads claude-relay-error-<date>.log from here (absolute path, no ~ expansion) for its optional lead signal. Defaults to <plugin-data-dir>/account-rotation/crs-logs if unset — point it at your CRS deployment's actual log directory. logTzOffset must match the timezone your CRS instance timestamps logs in.",+ "_logDir_note": "Optional: crs-429-cooldown.mjs reads claude-relay-error-<date>.log from here (absolute path, no ~ expansion) for its optional lead signal. Defaults to <plugin-data-dir>/account-rotation/crs-logs if unset — point it at your CLIProxyAPI/legacy relay deployment's actual log directory. logTzOffset must match the timezone your relay instance timestamps logs in.","errorLogWindowMs": 300000,"errorLogCooldownMs": 300000,"_errorLog_note": "errorLogWindowMs: how far back crs-429-cooldown.mjs scans the log for a 429 hit. errorLogCooldownMs: how long a log-only hit holds an account before rateLimitStatus itself would (default 5m each).",@@ -62,9 +62,9 @@"_cooldownStatePath_note": "Optional: path to a separate 429/rate-limit reconciler's state file, read-only, so crs-401-refresher.mjs's re-enable sweep doesn't fight a genuine rate-limit window. Leave null if you don't run one — treated as 'no known windows', never an error. NEVER point this at the same file crs-401-refresher.mjs writes (crs-401-state.json) — each reconciler must own its own state file.","manualDisableEnabled": true,"cooldownEnabled": false,- "_cooldownEnabled_note": "Off by default (or $CRS_COOLDOWN_ENABLED=1). Turns on crs-429-cooldown.mjs, a fast (1min-cadence) reconciler that holds an account out of rotation only on a real rateLimitStatus.isRateLimited=true from CRS (plus the optional error-log lead signal under logDir/logTzOffset above) — never on utilization. Own state file at <stateDir>/crs-429-state.json.",+ "_cooldownEnabled_note": "Off by default (or $CRS_COOLDOWN_ENABLED=1). Turns on crs-429-cooldown.mjs, a fast (1min-cadence) reconciler that holds an account out of rotation only on a real rateLimitStatus.isRateLimited=true from the CLIProxyAPI/legacy relay (plus the optional error-log lead signal under logDir/logTzOffset above) — never on utilization. Own state file at <stateDir>/crs-429-state.json.","tokenRefreshEnabled": false,- "_tokenRefreshEnabled_note": "Off by default (or $CRS_TOKEN_REFRESH_ENABLED=1). Turns on crs-401-refresher.mjs, which proactively refreshes each CRS account's OAuth access token via POST /admin/claude-accounts/:id/refresh before it expires, reviving sidelined accounts on success and flagging a dead refresh_token as needs-reauth on failure. Own state file at <stateDir>/crs-401-state.json. Before reviving an account it checks cooldownStatePath (read-only) so it never re-enables one crs-429-cooldown.mjs is genuinely holding for a live rate limit.",+ "_tokenRefreshEnabled_note": "Off by default (or $CRS_TOKEN_REFRESH_ENABLED=1). Turns on crs-401-refresher.mjs, which proactively refreshes each CLIProxyAPI/legacy relay account's OAuth access token via POST /admin/claude-accounts/:id/refresh before it expires, reviving sidelined accounts on success and flagging a dead refresh_token as needs-reauth on failure. Own state file at <stateDir>/crs-401-state.json. Before reviving an account it checks cooldownStatePath (read-only) so it never re-enables one crs-429-cooldown.mjs is genuinely holding for a live rate limit.","tokenRefreshWindowMs": 1800000,"_tokenRefreshWindowMs_note": "Refresh an account's token once it's within this many ms of expiry (default 30m).","enableMagicLinkRecovery": false,diff --git a/claude-ops/scripts/account-rotation/crs-429-cooldown.mjs b/claude-ops/scripts/account-rotation/crs-429-cooldown.mjsindex b36f3127..33664ee0 100644--- a/claude-ops/scripts/account-rotation/crs-429-cooldown.mjs+++ b/claude-ops/scripts/account-rotation/crs-429-cooldown.mjs@@ -4,18 +4,18 @@// OPT-IN. Does nothing (exits 0) unless crs.cooldownEnabled is true (or// $CRS_COOLDOWN_ENABLED=1) — installing this plugin never silently starts a new// daemon. Enable it alongside crs-priority-daemon.mjs when you run a self-hosted-// claude-relay-service (CRS) pool and want a fast, real-signal-only cooldown watcher+// CLIProxyAPI/legacy CRS-compatible relay pool and want a fast, real-signal-only cooldown watcher// that runs on a tighter cadence than the priority daemon.//-// Two signals, in priority order — both are things Anthropic (or CRS observing+// Two signals, in priority order — both are things Anthropic (or the relay observing// Anthropic) actually told us, never a speculative/utilization-based guess:-// 1. rateLimitStatus.isRateLimited = true on the account record (CRS derived-// this from a real upstream response on an actual inference call).-// 2. (optional) tail of today's CRS relay error log for 429 lines mentioning the-// account name — a faster lead signal than #1 while CRS's own record catches-// up. Reads from crs.logDir (or $CRS_LOG_DIR); a no-op on a fresh install+// 1. rateLimitStatus.isRateLimited = true on the account record (relay-derived+// from a real upstream response on an actual inference call).+// 2. (optional) tail of today's relay error log for 429 lines mentioning the+// account name — a faster lead signal than #1 while the relay's own record catches+// up. Reads from crs.logDir (or $CRS_LOG_DIR; legacy names); a no-op on a fresh install// since that directory won't exist yet (this reconciler never assumes a-// real CRS deployment's log layout).+// real relay deployment's log layout).//// State is owned exclusively by THIS reconciler (own file, atomic write, single-// flight lock via crs-reconciler-state.mjs) so that:diff --git a/claude-ops/scripts/account-rotation/crs-pool-config.mjs b/claude-ops/scripts/account-rotation/crs-pool-config.mjsindex e869fcce..52445194 100644--- a/claude-ops/scripts/account-rotation/crs-pool-config.mjs+++ b/claude-ops/scripts/account-rotation/crs-pool-config.mjs@@ -1,9 +1,9 @@#!/usr/bin/env node/**- * crs-pool-config.mjs — shared CRS + rotation config resolution for public installs.+ * crs-pool-config.mjs — shared CLIProxyAPI/legacy CRS-compatible relay + rotation config resolution for public installs.*- * Vault key ↔ CRS account name mapping is NEVER hardcoded in repo scripts.- * Configure per account via `crsAccountName` and/or optional `crs.nameByVaultKey`.+ * Vault key ↔ legacy relay account name mapping is NEVER hardcoded in repo scripts.+ * Configure per account via legacy fields `crsAccountName` and/or optional `crs.nameByVaultKey`.*/import { readFileSync, existsSync } from 'fs';@@ -134,8 +134,8 @@ export function resolveCredentialStorePath() {}/**- * Resolve the CRS admin password for HTTP basic auth against a self-hosted- * claude-relay-service pool. Resolution order (first hit wins), none of it+ * Resolve the relay admin password for HTTP basic auth against a self-hosted+ * CLIProxyAPI/legacy CRS-compatible pool. Resolution order (first hit wins), none of it* ever hardcoded to one deployment:* 1. $CRS_ADMIN_PASSWORD* 2. the env var named by config.crs.adminPasswordEnv (or $CRS_ADMIN_PASSWORD_ENV)diff --git a/claude-ops/scripts/account-rotation/crs-priority-daemon.mjs b/claude-ops/scripts/account-rotation/crs-priority-daemon.mjsindex 815bd222..aa9f6b25 100755--- a/claude-ops/scripts/account-rotation/crs-priority-daemon.mjs+++ b/claude-ops/scripts/account-rotation/crs-priority-daemon.mjs@@ -1,9 +1,9 @@#!/usr/bin/env node// crs-priority-daemon.mjs — utilization-driven account prioritization for a-// claude-relay-service (CRS) pool.+// CLIProxyAPI/legacy CRS-compatible relay pool.//-// CRS load-balances Claude requests across many claude.ai accounts at once. Each-// account carries a `schedulable` flag; CRS only routes to schedulable accounts.+// The relay load-balances Claude requests across many claude.ai accounts at once. Each+// account carries a `schedulable` flag; the relay only routes to schedulable accounts.// This daemon flips that flag from LIVE signals so the pool AVOIDS near-maxed// accounts (cutting 429/529s) and RE-ENABLES them once their windows recover —// the relay-pool analogue of the keychain rotator's "rotate to the coolestdiff --git a/claude-ops/scripts/account-rotation/crs-reconciler-state.mjs b/claude-ops/scripts/account-rotation/crs-reconciler-state.mjsindex 1ed92480..7128e7ea 100644--- a/claude-ops/scripts/account-rotation/crs-reconciler-state.mjs+++ b/claude-ops/scripts/account-rotation/crs-reconciler-state.mjs@@ -1,8 +1,8 @@#!/usr/bin/env node/**- * crs-reconciler-state.mjs — shared own-state-file helper for CRS pool reconcilers.+ * crs-reconciler-state.mjs — shared own-state-file helper for CLIProxyAPI/legacy CRS-compatible pool reconcilers.*- * Each reconciler that watches a claude-relay-service (CRS) account pool (cooldown+ * Each reconciler that watches a relay account pool (cooldown* tracking, token-refresh tracking, priority scoring, ...) MUST persist to its OWN* state file — never a file shared with another reconciler. Sharing one file was the* root cause of a documented production incident: reconciler A's non-atomic writediff --git a/claude-ops/scripts/account-rotation/ops-accounts-backend.mjs b/claude-ops/scripts/account-rotation/ops-accounts-backend.mjsindex 04226585..1fc5a234 100644--- a/claude-ops/scripts/account-rotation/ops-accounts-backend.mjs+++ b/claude-ops/scripts/account-rotation/ops-accounts-backend.mjs@@ -1,13 +1,13 @@/**- * ops-accounts-backend.mjs — resolve CRS vs local seat-state backend.+ * ops-accounts-backend.mjs — resolve legacy CRS-compatible relay vs local seat-state backend.** Env / config:- * OPS_ACCOUNTS_BACKEND | crs.backend- * local | seat-state | file → local seat-state (no CRS)- * crs → CRS admin API only- * auto (default) → try CRS, fall back to local+ * OPS_ACCOUNTS_BACKEND | crs.backend (legacy config key)+ * local | seat-state | file → local seat-state (no relay)+ * crs → legacy CRS-compatible admin API only+ * auto (default) → try relay, fall back to local*- * Dual-write (CRS path → seat-state.json) is on by default unless+ * Dual-write (relay path → seat-state.json) is on by default unless* OPS_ACCOUNTS_DUAL_WRITE=0.*/export function resolveAccountsBackend({ env = process.env, cfgBackend } = {}) {diff --git a/claude-ops/scripts/account-rotation/refresh-tokens.mjs b/claude-ops/scripts/account-rotation/refresh-tokens.mjsindex acde1fec..32c8d4a7 100644--- a/claude-ops/scripts/account-rotation/refresh-tokens.mjs+++ b/claude-ops/scripts/account-rotation/refresh-tokens.mjs@@ -40,7 +40,7 @@ const LOG_PATH = join(__dirname, 'rotation.log');const NEEDS_REAUTH_PATH = join(__dirname, '.crs-token-refresher-state.json');const AUTOLOOP_STATE_PATH = join(__dirname, '.crs-magic-autoloop-state.json');const KEYCHAIN_SERVICE = 'Claude Code-credentials';-const KEYCHAIN_ACCOUNT = process.env.USER || '<USER>';+const KEYCHAIN_ACCOUNT = process.env.USER || 'claude-ops';const CLIENT_ID = '9d1c250a-e61b-44d9-88ed-5944d1962f5e';const TOKEN_ENDPOINT = 'https://platform.claude.com/v1/oauth/token';diff --git a/claude-ops/scripts/account-rotation/use-bedrock.sh b/claude-ops/scripts/account-rotation/use-bedrock.shindex b86975e0..7a7f697e 100755--- a/claude-ops/scripts/account-rotation/use-bedrock.sh+++ b/claude-ops/scripts/account-rotation/use-bedrock.sh@@ -18,7 +18,7 @@ _ROT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"if [[ "${CLAUDE_CONFIRM_METERED_BEDROCK:-}" != "1" ]]; thenecho "⛔ Bedrock fallback blocked."- echo " Bedrock is metered AWS usage. OAuth/CRS is the preferred route."+ echo " Bedrock is metered AWS usage. OAuth via CLIProxyAPI/legacy relay is the preferred route."echo " Reason OAuth is unavailable: ${CLAUDE_BEDROCK_REASON:-not supplied}"echo " To confirm for this short session: CLAUDE_CONFIRM_METERED_BEDROCK=1 source ~/.claude/scripts/account-rotation/use-bedrock.sh"return 2 2>/dev/null || exit 2diff --git a/claude-ops/scripts/account-rotation/use-oauth.sh b/claude-ops/scripts/account-rotation/use-oauth.shindex d036a943..13726c12 100755--- a/claude-ops/scripts/account-rotation/use-oauth.sh+++ b/claude-ops/scripts/account-rotation/use-oauth.sh@@ -1,9 +1,9 @@#!/bin/bash-# use-oauth.sh — exit Bedrock mode in this shell, restore CRS-backed OAuth defaults.+# use-oauth.sh — exit Bedrock mode in this shell, restore CLIProxyAPI/legacy CRS-backed OAuth defaults.# Source me: `source ~/.claude/scripts/account-rotation/use-oauth.sh`## Persists to ~/.claude/settings.json via claude-routing-state:-# ANTHROPIC_BASE_URL points at CRS and CLAUDE_CODE_OAUTH_TOKEN is the CRS relay token.+# ANTHROPIC_BASE_URL points at the relay and CLAUDE_CODE_OAUTH_TOKEN is the CLIProxyAPI-compatible relay token.unset CLAUDE_CODE_USE_BEDROCKunset AWS_BEDROCK_REGION@@ -23,7 +23,7 @@ _ROT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"ACTIVE=$(python3 -c "import json; print(json.load(open('$HOME/.claude/scripts/account-rotation/state.json')).get('activeAccount',''))" 2>/dev/null)echo "✅ OAuth mode restored for this shell + persisted to settings.json"echo " models : (subscription catalog — no hardcoded IDs)"-echo " route : CRS OAuth"+echo " route : CLIProxyAPI/legacy relay OAuth"echo " active account: ${ACTIVE:-(unknown)}"if [[ -f ~/.claude/.bedrock-fallback.json ]]; thenecho " ⚠ Bedrock sentinel still present — clearing"diff --git a/claude-ops/scripts/crsproxy-reauth/account_policy.example.yaml b/claude-ops/scripts/crsproxy-reauth/account_policy.example.yamlindex 5ea82859..324ac057 100644--- a/claude-ops/scripts/crsproxy-reauth/account_policy.example.yaml+++ b/claude-ops/scripts/crsproxy-reauth/account_policy.example.yaml@@ -1,9 +1,8 @@-# CRSProxy Account Policy — Example Template (NO PII)+# cliproxy Account Policy — Example Template (NO PII)## This is a PII-free example of the declarative account policy file.-# The real file is deployed to /opt/crsproxy/account_policy.yaml on the hub-# only — it contains real account emails and must NEVER be committed to the-# public repo.+# The real file is deployed to the cliproxy hub only — it contains real account+# emails and must NEVER be committed to the public repo.## The auto_reauth.py monitor reads this file to reconcile actual auth files# against the declared target state:@@ -18,9 +17,16 @@# priority: lower = higher priority for routing (1=highest)# profile_id: Browser Use Cloud profile UUID (optional, for profile-based reauth)# reauth_method: profile | email | x_oauth | none-# requires_2fa: true if the account needs 2FA (human checkpoint required)+# two_factor: optional 2FA integration/checkpoint config# x_username: X/Twitter username (optional, for x_oauth method only)# max_consecutive_failures: auto-disable after this many consecutive failures+#+# 2FA methods:+# none no 2FA expected+# onepassword_totp fetch TOTP seed/URI from local 1Password Connect, send only the short-lived code to Browser Use+# passkey_checkpoint keep Browser Use session alive for human passkey/WebAuthn approval+# sms_checkpoint ask the operator for an SMS code and have Browser Use enter it+# push_checkpoint ask the operator to approve a push prompt, then resumeaccounts:# --- Claude OAuth accounts ---@@ -29,7 +35,8 @@ accounts:enabled: truepriority: 5reauth_method: email- requires_2fa: false+ two_factor:+ method: nonemax_consecutive_failures: 3- provider: claude@@ -38,7 +45,11 @@ accounts:priority: 3profile_id: "<BU_PROFILE_UUID>"reauth_method: profile- requires_2fa: false+ two_factor:+ method: onepassword_totp+ op_vault_id: "<OP_VAULT_UUID>"+ op_item_id: "<OP_ITEM_UUID>"+ op_field: "one-time password"max_consecutive_failures: 3- provider: claude@@ -46,7 +57,8 @@ accounts:enabled: truepriority: 5reauth_method: email- requires_2fa: false+ two_factor:+ method: sms_checkpointmax_consecutive_failures: 3# --- Codex (OpenAI) OAuth accounts ---@@ -55,7 +67,8 @@ accounts:enabled: truepriority: 7reauth_method: email- requires_2fa: false+ two_factor:+ method: nonemax_consecutive_failures: 3- provider: codex@@ -63,7 +76,8 @@ accounts:enabled: truepriority: 7reauth_method: email- requires_2fa: false+ two_factor:+ method: passkey_checkpointmax_consecutive_failures: 3# --- xAI/Grok OAuth accounts ---@@ -72,7 +86,8 @@ accounts:enabled: truepriority: 8reauth_method: email- requires_2fa: false+ two_factor:+ method: nonemax_consecutive_failures: 3- provider: xai@@ -81,7 +96,8 @@ accounts:reason: "different tier, no SuperGrok"priority: 8reauth_method: email- requires_2fa: false+ two_factor:+ method: nonemax_consecutive_failures: 3- provider: xai@@ -89,7 +105,8 @@ accounts:enabled: truepriority: 8reauth_method: x_oauth- requires_2fa: true+ two_factor:+ method: push_checkpointx_username: "example_user"max_consecutive_failures: 3@@ -100,5 +117,6 @@ accounts:reason: "Anthropic models should route via Claude OAuth, not Google Vertex AI"priority: 9reauth_method: none- requires_2fa: false+ two_factor:+ method: nonemax_consecutive_failures: 3diff --git a/claude-ops/scripts/crsproxy-reauth/auto_reauth.py b/claude-ops/scripts/crsproxy-reauth/auto_reauth.pyindex 0297f31f..90e530f9 100644--- a/claude-ops/scripts/crsproxy-reauth/auto_reauth.py+++ b/claude-ops/scripts/crsproxy-reauth/auto_reauth.py@@ -1,23 +1,23 @@#!/usr/bin/env python3-"""CRSProxy autonomous reauth monitor.+"""cliproxy autonomous reauth monitor.-Runs periodically via the systemd timer ``crsproxy-reauth.timer``.+Runs periodically via the cliproxy reauth systemd timer.Responsibilities:1. Run ``cliproxy-pool-health`` and log the result.2. Reconcile actual auth files against the declarative account policy- (``/opt/crsproxy/account_policy.yaml``):+ on the cliproxy hub:- If policy says enabled but auth file is disabled -> trigger reauth.- If policy says disabled but auth file is enabled -> disable it.3. Scan auth files for expired or disabled accounts.4. Trigger reauth for any expired account using **profile-first** logic:- If ``reauth_seats.json`` maps a Browser Use Cloud profile ID to theaccount -> invoke ``bu_profile_reauth.py`` with ``-profile-id``.- - Otherwise -> fall back to ``bu_reauth_lib.py`` (email-based).+ - Otherwise -> fall back to ``bu_reauth.py`` (email-based).A per-account cooldown prevents hammering a persistently-blocked account.5. Log all results WITHOUT exposing secrets (emails are masked, tokens /codes / cookies / callback URLs are never printed).- 6. NEVER restart, reload, or kill ``crsproxy.service`` — this monitor is+ 6. NEVER restart, reload, or kill the cliproxy service — this monitor isstrictly read-only with respect to the proxy process. Auth fileshot-reload, so reauth never requires a service restart.@@ -31,6 +31,8 @@ Exit codes:exit — only monitor-level breakage does."""+from __future__ import annotations+import jsonimport osimport re@@ -43,10 +45,12 @@ from pathlib import Path# ---------------------------------------------------------------------------# Configuration# ---------------------------------------------------------------------------+# Runtime compatibility: the deployed cliproxy hub may still use this legacy+# filesystem path while service paths are migrated.AUTH_DIR = Path("/opt/crsproxy/auths")POOL_HEALTH_CMD = "/usr/local/bin/cliproxy-pool-health"REAUTH_CMD = "/opt/crsproxy/venv/bin/python"-REAUTH_SCRIPT = "/opt/crsproxy/bu_reauth_lib.py"+REAUTH_SCRIPT = "/opt/crsproxy/bu_reauth.py"PROFILE_REAUTH_SCRIPT = "/opt/crsproxy/bu_profile_reauth.py"ENV_FILE = Path("/opt/crsproxy/.env")STATE_FILE = Path("/opt/crsproxy/.auto_reauth-state.json")@@ -118,8 +122,8 @@ def mask_email(email: str) -> str:# ----------------------------------------------------------------------------# Environment loading — read /opt/crsproxy/.env so the reauth subprocess-# inherits BROWSER_USE_API_KEY and Gmail/gogcli credentials.+# Environment loading — read the legacy deployed runtime .env so the reauth+# subprocess inherits BROWSER_USE_API_KEY and Gmail/gogcli credentials.# ---------------------------------------------------------------------------def load_env(path: Path) -> dict:"""Parse a simple KEY=VALUE .env file into a dict. Values may be wrapped@@ -264,6 +268,16 @@ def find_seat(seats: dict, provider: str, email: str) -> dict | None:return None+def find_policy_account(policy: dict | None, provider: str, email: str) -> dict | None:+ """Find the account-policy entry matching provider+email."""+ if not policy:+ return None+ for entry in policy.get("accounts", []):+ if entry.get("provider") == provider and entry.get("email") == email:+ return entry+ return None++# ---------------------------------------------------------------------------# Auth file lookup# ---------------------------------------------------------------------------@@ -459,7 +473,7 @@ def trigger_reauth(account: dict, env: dict,If ``reauth_seats.json`` maps a Browser Use Cloud profile ID to thisaccount, invoke ``bu_profile_reauth.py`` with ``-profile-id``. Otherwise- fall back to ``bu_reauth_lib.py`` (email-based).+ fall back to ``bu_reauth.py`` (email-based).Secrets are never logged: emails are masked, and the subprocess inheritsthe loaded .env so BROWSER_USE_API_KEY and Gmail credentials are@@ -474,6 +488,18 @@ def trigger_reauth(account: dict, env: dict,if seats is not None:seat = find_seat(seats, provider, email)+ two_factor = account.get("two_factor") or {}+ two_factor_method = str(two_factor.get("method") or "none").strip()+ two_factor_args = []+ if two_factor_method and two_factor_method != "none":+ two_factor_args.extend(["-two-factor-method", two_factor_method])+ if two_factor_method == "onepassword_totp":+ two_factor_args.extend([+ "-op-vault-id", str(two_factor.get("op_vault_id") or ""),+ "-op-item-id", str(two_factor.get("op_item_id") or ""),+ "-op-field", str(two_factor.get("op_field") or "one-time password"),+ ])+if seat and seat.get("profile_id"):# --- Profile-based reauth ---profile_id = seat["profile_id"]@@ -482,10 +508,12 @@ def trigger_reauth(account: dict, env: dict,"-provider", provider,"-email", email,"-profile-id", profile_id,+ *two_factor_args,]method = "profile"log(f"[reauth] triggering {provider} for {masked} via {method} "- f"(profile={profile_id[:12]}... reason: {account['reason']})")+ f"(profile={profile_id[:12]}... 2fa={two_factor_method} "+ f"reason: {account['reason']})")else:# --- Email-based reauth (fallback) ---# Claude verification emails are forwarded to a central Gmail inbox@@ -507,10 +535,11 @@ def trigger_reauth(account: dict, env: dict,"-provider", provider,"-email", email,"-gog-account", gog_account,+ *two_factor_args,]method = "email"log(f"[reauth] triggering {provider} for {masked} via {method} "- f"(reason: {account['reason']})")+ f"(2fa={two_factor_method} reason: {account['reason']})")try:proc = subprocess.run(@@ -610,6 +639,7 @@ def reconcile_accounts(env: dict, seats: dict, state: dict) -> dict:"provider": provider,"email": email,"reason": "policy-enabled-but-disabled",+ "two_factor": entry.get("two_factor") or {},}success, reason = trigger_reauth(account, env, seats)record_attempt(state, key, success, reason)@@ -720,6 +750,11 @@ def main() -> int:f"in cooldown — last attempt too recent")continue+ policy_entry = find_policy_account(+ policy, account["provider"], account["email"])+ if policy_entry and "two_factor" in policy_entry:+ account = {**account, "two_factor": policy_entry.get("two_factor") or {}}+attempted += 1success, reason = trigger_reauth(account, env, seats)record_attempt(state, key, success, reason)diff --git a/claude-ops/scripts/crsproxy-reauth/bu_2fa.py b/claude-ops/scripts/crsproxy-reauth/bu_2fa.pynew file mode 100644index 00000000..6a85852f--- /dev/null+++ b/claude-ops/scripts/crsproxy-reauth/bu_2fa.py@@ -0,0 +1,209 @@+#!/usr/bin/env python3+"""cliproxy Browser Use 2FA helpers.++This module supports the Browser Use reauth scripts without exposing long-lived+secrets to Browser Use Cloud. It reads 1Password Connect configuration from the+local runtime environment, fetches a configured OTP field, and returns only the+current short-lived TOTP code to the caller.+"""++from __future__ import annotations++import base64+import hashlib+import hmac+import json+import os+import re+import struct+import time+from pathlib import Path+from urllib.parse import parse_qs, urlparse++# Runtime compatibility: the deployed hub may still store cliproxy files under+# this legacy path while service paths are migrated. Do not treat the path name+# as product terminology.+LEGACY_RUNTIME_HOME = Path(os.environ.get("CLIPROXY_HOME", "/opt/crsproxy"))+DEFAULT_ENV_FILE = LEGACY_RUNTIME_HOME / ".env"+DEFAULT_OP_FIELD = "one-time password"+++class TwoFactorError(RuntimeError):+ """Raised when a configured 2FA method cannot produce a code."""+++def load_env_file(path: Path = DEFAULT_ENV_FILE) -> dict[str, str]:+ """Read a simple KEY=VALUE env file without logging any values."""+ env: dict[str, str] = {}+ try:+ text = path.read_text()+ except FileNotFoundError:+ return env+ except OSError as exc:+ raise TwoFactorError(f"cannot read env file: {exc.__class__.__name__}")++ for raw in text.splitlines():+ line = raw.strip()+ if not line or line.startswith("#") or "=" not in line:+ continue+ key, _, value = line.partition("=")+ key = key.strip()+ value = value.strip()+ if len(value) >= 2 and value[0] == value[-1] and value[0] in ("'", '"'):+ value = value[1:-1]+ env[key] = value+ return env+++def merged_runtime_env(env: dict[str, str] | None = None,+ env_file: Path = DEFAULT_ENV_FILE) -> dict[str, str]:+ """Merge os.environ with the runtime env file, preferring os.environ."""+ merged = load_env_file(env_file)+ merged.update(os.environ)+ if env:+ merged.update(env)+ return merged+++def _op_connect_settings(env: dict[str, str]) -> tuple[str, str]:+ host = env.get("OP_CONNECT_HOST") or env.get("OP_CONNECT_URL") or ""+ token = env.get("OP_CONNECT_TOKEN") or ""+ if not host:+ raise TwoFactorError("OP_CONNECT_HOST is not configured")+ if not token:+ raise TwoFactorError("OP_CONNECT_TOKEN is not configured")+ return host.rstrip("/"), token+++def fetch_1password_item(vault_id: str, item_id: str,+ env: dict[str, str] | None = None,+ timeout: int = 15) -> dict:+ """Fetch one item from 1Password Connect using local credentials."""+ if not vault_id or not item_id:+ raise TwoFactorError("1Password vault/item id missing")+ try:+ import requests+ except ImportError as exc:+ raise TwoFactorError("Python requests package is required for 1Password Connect") from exc+ host, token = _op_connect_settings(merged_runtime_env(env))+ url = f"{host}/v1/vaults/{vault_id}/items/{item_id}"+ try:+ response = requests.get(+ url,+ headers={"Authorization": f"Bearer {token}"},+ timeout=timeout,+ )+ except requests.RequestException as exc:+ raise TwoFactorError(f"1Password Connect request failed: {exc.__class__.__name__}")+ if response.status_code == 401:+ raise TwoFactorError("1Password Connect rejected the token")+ if response.status_code == 404:+ raise TwoFactorError("1Password item or vault not found")+ try:+ response.raise_for_status()+ except requests.HTTPError as exc:+ raise TwoFactorError(f"1Password Connect HTTP {response.status_code}") from exc+ try:+ return response.json()+ except json.JSONDecodeError as exc:+ raise TwoFactorError("1Password Connect returned invalid JSON") from exc+++def item_field_value(item: dict, field_name: str = DEFAULT_OP_FIELD) -> str:+ """Return a field value by label/id/type from a 1Password item."""+ wanted = (field_name or DEFAULT_OP_FIELD).strip().lower()+ candidates = []+ for field in item.get("fields", []) or []:+ if not isinstance(field, dict):+ continue+ label = str(field.get("label", "")).strip().lower()+ field_id = str(field.get("id", "")).strip().lower()+ field_type = str(field.get("type", "")).strip().lower()+ value = field.get("value") or field.get("totp") or ""+ candidates.append((label, field_id, field_type, str(value)))++ for label, field_id, field_type, value in candidates:+ if value and wanted in {label, field_id}:+ return value+ for label, field_id, field_type, value in candidates:+ if value and ("otp" in label or "otp" in field_id or "otp" in field_type):+ return value+ raise TwoFactorError(f"1Password OTP field not found: {field_name}")+++def _normalize_base32(secret: str) -> str:+ cleaned = re.sub(r"\s+", "", secret).upper()+ padding = "=" * ((8 - len(cleaned) % 8) % 8)+ return cleaned + padding+++def totp_from_secret(secret: str, now: int | None = None,+ period: int = 30, digits: int = 6,+ algorithm: str = "SHA1") -> str:+ """Generate a TOTP code using only the Python standard library."""+ if not secret:+ raise TwoFactorError("empty TOTP secret")+ if digits <= 0 or digits > 10:+ raise TwoFactorError("unsupported TOTP digit count")+ algo = algorithm.upper().replace("-", "")+ digest = {+ "SHA1": hashlib.sha1,+ "SHA256": hashlib.sha256,+ "SHA512": hashlib.sha512,+ }.get(algo)+ if digest is None:+ raise TwoFactorError(f"unsupported TOTP algorithm: {algorithm}")+ try:+ key = base64.b32decode(_normalize_base32(secret), casefold=True)+ except Exception as exc:+ raise TwoFactorError("invalid TOTP secret encoding") from exc+ counter = int((now if now is not None else time.time()) // period)+ msg = struct.pack(">Q", counter)+ mac = hmac.new(key, msg, digest).digest()+ offset = mac[-1] & 0x0F+ code_int = struct.unpack(">I", mac[offset:offset + 4])[0] & 0x7FFFFFFF+ return str(code_int % (10 ** digits)).zfill(digits)+++def totp_from_otpauth(value: str, now: int | None = None) -> str:+ """Generate a TOTP from an otpauth:// URI or raw base32 seed."""+ value = value.strip()+ if value.startswith("otpauth://"):+ parsed = urlparse(value)+ params = parse_qs(parsed.query)+ secret = (params.get("secret") or [""])[0]+ period = int((params.get("period") or ["30"])[0])+ digits = int((params.get("digits") or ["6"])[0])+ algorithm = (params.get("algorithm") or ["SHA1"])[0]+ return totp_from_secret(secret, now=now, period=period,+ digits=digits, algorithm=algorithm)+ return totp_from_secret(value, now=now)+++def current_totp_from_1password(vault_id: str, item_id: str,+ field_name: str = DEFAULT_OP_FIELD,+ env: dict[str, str] | None = None,+ now: int | None = None) -> str:+ """Fetch a configured 1Password OTP field and return the current code."""+ item = fetch_1password_item(vault_id, item_id, env=env)+ otp_value = item_field_value(item, field_name)+ return totp_from_otpauth(otp_value, now=now)+++def current_totp_from_config(config: dict | None,+ env: dict[str, str] | None = None,+ now: int | None = None) -> str | None:+ """Return a code for a supported two_factor config, or None if disabled."""+ config = config or {}+ method = str(config.get("method") or "none").strip().lower()+ if method in ("", "none", "false"):+ return None+ if method != "onepassword_totp":+ raise TwoFactorError(f"2FA method requires human checkpoint: {method}")+ return current_totp_from_1password(+ vault_id=str(config.get("op_vault_id") or ""),+ item_id=str(config.get("op_item_id") or ""),+ field_name=str(config.get("op_field") or DEFAULT_OP_FIELD),+ env=env,+ now=now,+ )diff --git a/claude-ops/scripts/crsproxy-reauth/bu_profile_reauth.py b/claude-ops/scripts/crsproxy-reauth/bu_profile_reauth.pyindex a7ab25fc..ab92472d 100644--- a/claude-ops/scripts/crsproxy-reauth/bu_profile_reauth.py+++ b/claude-ops/scripts/crsproxy-reauth/bu_profile_reauth.py@@ -1,5 +1,5 @@#!/usr/bin/env python3-"""Browser Use Cloud — Profile-based OAuth re-auth for CRSProxy.+"""Browser Use Cloud — profile-based OAuth re-auth for cliproxy / CLIProxyAPI.Uses a Browser Use Cloud persistent profile (with synced cookies) tobypass the OAuth email/captcha flow. The agent starts already@@ -14,6 +14,8 @@ Usage:-provider claude -email [email protected] -profile-id <BU_PROFILE_ID>"""+from __future__ import annotations+import argparseimport jsonimport os@@ -25,8 +27,8 @@ import timefrom datetime import datetime, timedelta, timezonefrom pathlib import Path-# Import infrastructure from bu_reauth_lib.py (the stable copy that avoids-# the quarantine issue affecting bu_reauth.py on the hub).+# Import deployed hub infrastructure from the stable runtime copy. The helper+# filename is legacy compatibility; current product wording is cliproxy/CLIProxyAPI.sys.path.insert(0, "/opt/crsproxy")from bu_reauth_lib import (BrowserUseClient,@@ -58,13 +60,80 @@ from bu_reauth_lib import (_timeout_handler,)-# Also import _complete_reauth for the final step+# Also import _complete_reauth and 2FA helpers for the final stepsfrom bu_reauth_lib import _complete_reauth+from bu_2fa import TwoFactorError, current_totp_from_config+++def needs_two_factor(result_text: str) -> bool:+ """Return True when the page appears to ask for a 2FA/OTP code."""+ text = result_text.lower()+ indicators = [+ "two-factor", "two factor", "2fa", "multi-factor", "mfa",+ "authenticator", "authentication code",+ "one-time password", "one time password",+ "enter a code from your authenticator",+ ]+ return any(indicator in text for indicator in indicators)+++def handle_two_factor_prompt(client: BrowserUseClient, session_id: str,+ result_text: str, two_factor: dict | None) -> bool:+ """Handle a 2FA prompt without exposing long-lived secrets to Browser Use."""+ if not needs_two_factor(result_text):+ return True++ config = two_factor or {}+ method = str(config.get("method") or "none").strip().lower()+ if method in ("", "none", "false"):+ log("[2FA] Prompt detected but no 2FA method is configured")+ return False++ if method == "onepassword_totp":+ try:+ code = current_totp_from_config(config)+ except TwoFactorError as exc:+ log(f"[2FA] Could not get 1Password TOTP: {exc}")+ return False+ if not code:+ log("[2FA] 1Password TOTP config returned no code")+ return False+ log("[2FA] 1Password TOTP generated (value redacted)")+ task = (+ f"Enter this one-time authentication code: {code}. "+ "Click Continue. Report exactly what you see."+ )+ else:+ log(f"[2FA] Human checkpoint required: {method}")+ if method == "sms_checkpoint":+ task = "Ask the operator for the SMS code, enter it, and click Continue."+ elif method == "push_checkpoint":+ task = "Ask the operator to approve the push prompt, then click Continue."+ elif method == "passkey_checkpoint":+ task = "Ask the operator to complete the passkey/WebAuthn prompt."+ else:+ task = "Ask the operator to complete the 2FA checkpoint."+ task += " Report exactly what you see."++ run = client.create_run(task, session_id=session_id)+ run_id = run.get("id", "")+ log(f"[2FA] Checkpoint run: {run_id[:12]}...")+ result = client.wait_for_run(run_id, timeout=RUN_POLL_TIMEOUT)+ text = str(result.get("result", "") or "")+ log(f"[2FA] Checkpoint complete: {text[:200]}")+ if detect_captcha(result, client):+ log("[2FA] Captcha appeared during 2FA checkpoint")+ return False+ if any(ind in text.lower() for ind in ["invalid", "incorrect", "expired", "try again"]):+ log("[2FA] 2FA code was rejected")+ return False+ return Truedef run_profile_reauth(provider: str, email: str, profile_id: str,client: BrowserUseClient,- dry_run: bool = False) -> int:+ dry_run: bool = False,+ two_factor: dict | None = None) -> int:"""Run the profile-based OAuth reauth flow.1. Acquire serialization lease.@@ -91,14 +160,15 @@ def run_profile_reauth(provider: str, email: str, profile_id: str,try:return _run_profile_reauth_inner(provider, email, profile_id,- client, dry_run, log_path)+ client, dry_run, log_path, two_factor)finally:release_lease()def _run_profile_reauth_inner(provider: str, email: str, profile_id: str,client: BrowserUseClient, dry_run: bool,- log_path: Path) -> int:+ log_path: Path,+ two_factor: dict | None = None) -> int:"""Inner profile reauth flow — assumes lease is already held."""meta = PROVIDERS[provider]port = meta["callback_port"]@@ -186,6 +256,9 @@ def _run_profile_reauth_inner(provider: str, email: str, profile_id: str,stale_backup, email, provider, meta)return EXIT_CAPTCHA+ if not handle_two_factor_prompt(client, session_id, text, two_factor):+ return EXIT_FAILURE+# --- Step 5: Extract callback URL ---log("[4] Searching for callback URL...")callback_url = extract_callback_url(result, client, provider)@@ -401,7 +474,7 @@ def main():global _log_fileparser = argparse.ArgumentParser(- description="Profile-based OAuth reauth for CRSProxy")+ description="Profile-based OAuth reauth for cliproxy / CLIProxyAPI")parser.add_argument("-provider", required=True,choices=["claude", "xai", "codex"],help="OAuth provider")@@ -411,6 +484,16 @@ def main():help="Browser Use Cloud profile ID")parser.add_argument("-dry-run", action="store_true",help="Validate setup without creating browser runs")+ parser.add_argument("-two-factor-method", default="none",+ choices=["none", "onepassword_totp", "passkey_checkpoint",+ "sms_checkpoint", "push_checkpoint"],+ help="2FA method to use if the login flow asks for one")+ parser.add_argument("-op-vault-id", default="",+ help="1Password Connect vault UUID for onepassword_totp")+ parser.add_argument("-op-item-id", default="",+ help="1Password Connect item UUID for onepassword_totp")+ parser.add_argument("-op-field", default="one-time password",+ help="1Password item field containing the TOTP seed/URI")args = parser.parse_args()log_path = LOG_DIR / f"bu_profile_reauth_{args.provider}_{args.email.split('@')[0]}.log"@@ -434,6 +517,13 @@ def main():log("[FATAL] BROWSER_USE_API_KEY not found")return EXIT_FAILURE+ two_factor = {+ "method": args.two_factor_method,+ "op_vault_id": args.op_vault_id,+ "op_item_id": args.op_item_id,+ "op_field": args.op_field,+ }+log(f"=== Profile Reauth: {args.provider} / {safe_email(args.email)} ===")log(f"=== Profile ID: {args.profile_id[:12]}... ===")@@ -448,6 +538,7 @@ def main():profile_id=args.profile_id,client=client,dry_run=args.dry_run,+ two_factor=two_factor,)except TimeoutError as e:log(f"[TIMEOUT] {e}")diff --git a/claude-ops/scripts/crsproxy-reauth/bu_reauth.py b/claude-ops/scripts/crsproxy-reauth/bu_reauth.pyindex beff3122..546a61db 100644--- a/claude-ops/scripts/crsproxy-reauth/bu_reauth.py+++ b/claude-ops/scripts/crsproxy-reauth/bu_reauth.py@@ -1,8 +1,8 @@#!/usr/bin/env python3-"""Browser Use Cloud Agent — OAuth re-auth for CRSProxy.+"""Browser Use Cloud Agent — OAuth re-auth for cliproxy / CLIProxyAPI.Production-grade reauth script that uses Browser Use Cloud to automate-OAuth flows for Claude, xAI, and Codex providers on the healify-hub EC2.+OAuth flows for Claude, xAI, and Codex providers on the cliproxy hub.Features:- Proper timeout handling on all Browser Use API calls (never hang).@@ -20,8 +20,8 @@ Features:keeps browser session alive, waits for a file-based trigger, thenresumes with a follow-up run in the same session to click Authorizeand capture the callback URL.- - Serialization lease: checks /opt/crsproxy/state/crsproxy-claude-oauth-lease.json- before starting a login, waits if held, cleans up after completion.+ - Serialization lease: checks the cliproxy OAuth lease file before starting+ a login, waits if held, cleans up after completion.- Email cooldown: tracks verification code send timestamps, enforcesa 5-minute cooldown if >3 codes are sent in 5 minutes, logscooldown events without exposing the email address or code values.@@ -65,6 +65,8 @@ Usage:# The script will then create a follow-up run and complete the flow."""+from __future__ import annotations+import argparseimport atexitimport json@@ -77,11 +79,14 @@ import timefrom datetime import datetime, timedelta, timezonefrom pathlib import Path-import requests+from bu_2fa import TwoFactorError, current_totp_from_config# ---------------------------------------------------------------------------# Paths / constants# ---------------------------------------------------------------------------+# Runtime compatibility: the deployed hub may still store cliproxy files under+# this legacy path while service paths are migrated. Keep these constants until+# the host is moved; do not present the old path name as the product name.HUB_BIN = "/opt/crsproxy/cli-proxy-api"HUB_CONFIG = "/opt/crsproxy/config.yaml"AUTH_DIR = Path("/opt/crsproxy/auths")@@ -90,7 +95,8 @@ BU_API_BASE = "https://api.browser-use.com/api/v4"BU_MODEL = "gpt-5.6-luna"LOG_DIR = Path("/tmp")STATE_DIR = Path("/opt/crsproxy/state")-LEASE_FILE = STATE_DIR / "crsproxy-claude-oauth-lease.json"+LEASE_FILE = STATE_DIR / "cliproxy-claude-oauth-lease.json"+LEGACY_LEASE_FILE = STATE_DIR / "crsproxy-claude-oauth-lease.json"CANARY_URL = "http://localhost:8319/v1/chat/completions"# Exit codes@@ -116,7 +122,8 @@ CHECKPOINT_POLL_INTERVAL = 5 # seconds between trigger pollsCHECKPOINT_TIMEOUT = 600 # 10 minutes for human to solve captcha# Email cooldown configuration-EMAIL_COOLDOWN_FILE = STATE_DIR / "crsproxy-email-cooldown.json"+EMAIL_COOLDOWN_FILE = STATE_DIR / "cliproxy-email-cooldown.json"+LEGACY_EMAIL_COOLDOWN_FILE = STATE_DIR / "crsproxy-email-cooldown.json"EMAIL_COOLDOWN_WINDOW = 300 # 5 min sliding window for counting code sendsEMAIL_COOLDOWN_THRESHOLD = 3 # max 3 code sends before cooldown triggersEMAIL_COOLDOWN_DURATION = 300 # 5 min cooldown when threshold exceeded@@ -218,6 +225,15 @@ def safe_email(email: str) -> str:return f"{local[0]}***@{domain}"+def _requests_module():+ """Import requests lazily so local logic tests can run without hub deps."""+ try:+ import requests+ return requests+ except ImportError as exc:+ raise RuntimeError("Python requests package is required for Browser Use API calls") from exc++# ---------------------------------------------------------------------------# Canary request (proxy health check)# ---------------------------------------------------------------------------@@ -230,6 +246,7 @@ def send_canary(model: str, timeout: int = 30) -> int:has no enabled accounts — the candidate might fix this)."""try:+ requests = _requests_module()r = requests.post(CANARY_URL,headers={"Authorization": "Bearer test"},@@ -380,9 +397,10 @@ def acquire_lease(provider: str, max_wait: int = 120) -> bool:t0 = time.time()first_check = Truewhile True:- if LEASE_FILE.exists():+ lease_path = LEASE_FILE if LEASE_FILE.exists() else LEGACY_LEASE_FILE+ if lease_path.exists():try:- data = json.loads(LEASE_FILE.read_text())+ data = json.loads(lease_path.read_text())age = time.time() - data.get("timestamp", 0)if age < LEASE_STALE_SECONDS:if first_check:@@ -397,10 +415,10 @@ def acquire_lease(provider: str, max_wait: int = 120) -> bool:continue# Stale lease — remove itlog(f"Stale lease ({int(age)}s old) — removing")- LEASE_FILE.unlink()+ lease_path.unlink()except (json.JSONDecodeError, OSError):try:- LEASE_FILE.unlink()+ lease_path.unlink()except OSError:pass# Lease file doesn't exist or was removed — acquire@@ -428,12 +446,13 @@ def release_lease():# Email cooldown tracking# ---------------------------------------------------------------------------def _load_cooldown_state() -> dict:- """Load email cooldown state from file."""- try:- if EMAIL_COOLDOWN_FILE.exists():- return json.loads(EMAIL_COOLDOWN_FILE.read_text())- except Exception:- pass+ """Load email cooldown state from the current or legacy file."""+ for path in (EMAIL_COOLDOWN_FILE, LEGACY_EMAIL_COOLDOWN_FILE):+ try:+ if path.exists():+ return json.loads(path.read_text())+ except Exception:+ passreturn {"send_timestamps": [], "cooldown_until": 0}@@ -555,10 +574,11 @@ class BrowserUseClient:self._runs: list[str] = []self.keep_browser_alive = False # set True during human checkpoint- def _request(self, method: str, path: str, **kwargs) -> requests.Response:+ def _request(self, method: str, path: str, **kwargs):"""Make an HTTP request with a guaranteed timeout."""kwargs.setdefault("timeout", API_TIMEOUT)url = f"{self.base}{path}"+ requests = _requests_module()r = requests.request(method, url, headers=self.headers, **kwargs)r.raise_for_status()return r@@ -898,6 +918,74 @@ def _code_entry_failed(result_text: str) -> bool:return any(ind in text_lower for ind in failure_indicators)+def needs_two_factor(result_text: str) -> bool:+ """Return True when the page appears to be asking for a 2FA/OTP code."""+ text = result_text.lower()+ indicators = [+ "two-factor", "two factor", "2fa", "multi-factor", "mfa",+ "authenticator", "authentication code",+ "one-time password", "one time password",+ "enter a code from your authenticator",+ ]+ return any(indicator in text for indicator in indicators)+++def _human_2fa_message(method: str) -> str:+ """Return safe Browser Use instructions for a human-driven 2FA method."""+ if method == "sms_checkpoint":+ return "Ask the operator for the SMS code, enter it, and click Continue."+ if method == "push_checkpoint":+ return "Ask the operator to approve the push prompt, then click Continue."+ if method == "passkey_checkpoint":+ return "Ask the operator to complete the passkey/WebAuthn prompt."+ return "Ask the operator to complete the 2FA checkpoint."+++def handle_two_factor_prompt(client: BrowserUseClient, session_id: str,+ result_text: str, two_factor: dict | None) -> bool:+ """Handle a 2FA prompt without exposing long-lived secrets to Browser Use."""+ if not needs_two_factor(result_text):+ return True++ config = two_factor or {}+ method = str(config.get("method") or "none").strip().lower()+ if method in ("", "none", "false"):+ log("[2FA] Prompt detected but no 2FA method is configured")+ return False++ if method == "onepassword_totp":+ try:+ code = current_totp_from_config(config)+ except TwoFactorError as exc:+ log(f"[2FA] Could not get 1Password TOTP: {exc}")+ return False+ if not code:+ log("[2FA] 1Password TOTP config returned no code")+ return False+ log("[2FA] 1Password TOTP generated (value redacted)")+ task = (+ f"Enter this one-time authentication code: {code}. "+ "Click Continue. Report exactly what you see."+ )+ else:+ log(f"[2FA] Human checkpoint required: {method}")+ task = _human_2fa_message(method) + " Report exactly what you see."++ run = client.create_run(task, session_id=session_id)+ run_id = run.get("id", "")+ log(f"[2FA] Checkpoint run: {run_id[:12]}...")+ result = client.wait_for_run(run_id, timeout=RUN_POLL_TIMEOUT)+ text = str(result.get("result", "") or "")+ log(f"[2FA] Checkpoint complete: {text[:200]}")+ if detect_captcha(result, client):+ log("[2FA] Captcha appeared during 2FA checkpoint")+ return False+ if _code_entry_failed(text):+ log("[2FA] 2FA code was rejected")+ return False+ return True++def detect_captcha(run_result: dict, client: BrowserUseClient) -> bool:"""Check if the run result or events indicate a real captcha challenge."""result_raw = str(run_result.get("result", "") or "")@@ -1204,7 +1292,8 @@ def _complete_reauth(proc: subprocess.Popen, callback_url: str,def run_reauth(provider: str, email: str, gog_account: str,- client: BrowserUseClient, dry_run: bool = False) -> int:+ client: BrowserUseClient, dry_run: bool = False,+ two_factor: dict | None = None) -> int:"""Run the full Browser Use Cloud OAuth reauth flow.Returns exit code: 0=success, 1=failure, 2=captcha.@@ -1221,14 +1310,15 @@ def run_reauth(provider: str, email: str, gog_account: str,try:return _run_reauth_inner(provider, email, gog_account, client,- dry_run, log_path)+ dry_run, log_path, two_factor)finally:release_lease()def _run_reauth_inner(provider: str, email: str, gog_account: str,client: BrowserUseClient, dry_run: bool,- log_path: Path) -> int:+ log_path: Path,+ two_factor: dict | None = None) -> int:"""Inner reauth flow — assumes lease is already held."""meta = PROVIDERS[provider]port = meta["callback_port"]@@ -1302,6 +1392,9 @@ def _run_reauth_inner(provider: str, email: str, gog_account: str,stale_backup, email, provider, meta)return EXIT_FAILURE+ if not handle_two_factor_prompt(client, session_id, text1, two_factor):+ return EXIT_FAILURE+# --- Step 5: Handle verification code or magic link ---needs_code = any(kw in text1.lower() for kw in["code", "link sent", "verify", "check your email",@@ -1350,6 +1443,9 @@ def _run_reauth_inner(provider: str, email: str, gog_account: str,email, provider, meta)return EXIT_FAILURE+ if not handle_two_factor_prompt(client, session_id, text2, two_factor):+ return EXIT_FAILURE+# Check if code entry failedif _code_entry_failed(text2):log(f"[4] Code entry failed on attempt {attempt+1}")@@ -1406,6 +1502,9 @@ def _run_reauth_inner(provider: str, email: str, gog_account: str,return _complete_reauth(proc, callback_url, auth_file,stale_backup, email, provider, meta)return EXIT_CAPTCHA++ if not handle_two_factor_prompt(client, session_id, text3, two_factor):+ return EXIT_FAILUREelse:log("[5] No magic link found — cannot continue")return EXIT_FAILURE@@ -1437,6 +1536,9 @@ def _run_reauth_inner(provider: str, email: str, gog_account: str,return _complete_reauth(proc, callback_url, auth_file,stale_backup, email, provider, meta)return EXIT_CAPTCHA++ if not handle_two_factor_prompt(client, session_id, text2, two_factor):+ return EXIT_FAILUREelse:log("[FAIL] No verification code or magic link received")return EXIT_FAILURE@@ -1468,6 +1570,9 @@ def _run_reauth_inner(provider: str, email: str, gog_account: str,stale_backup, email, provider, meta)return EXIT_FAILURE+ if not handle_two_factor_prompt(client, session_id, text_auth, two_factor):+ return EXIT_FAILURE+# --- Step 7: Extract callback URL ---log("[7] Searching for callback URL...")callback_url = extract_callback_url(result_auth, client, provider)@@ -1668,7 +1773,7 @@ def main():global _log_fileparser = argparse.ArgumentParser(- description="Browser Use Cloud OAuth reauth for CRSProxy"+ description="Browser Use Cloud OAuth reauth for cliproxy / CLIProxyAPI")parser.add_argument("-provider", required=True,choices=list(PROVIDERS.keys()),@@ -1699,6 +1804,16 @@ def main():"/opt/crsproxy/state/bu_reauth_checkpoint.json for the ""session_id, then creates a follow-up run to ""click Authorize and complete the flow.")+ parser.add_argument("-two-factor-method", default="none",+ choices=["none", "onepassword_totp", "passkey_checkpoint",+ "sms_checkpoint", "push_checkpoint"],+ help="2FA method to use if the login flow asks for one")+ parser.add_argument("-op-vault-id", default="",+ help="1Password Connect vault UUID for onepassword_totp")+ parser.add_argument("-op-item-id", default="",+ help="1Password Connect item UUID for onepassword_totp")+ parser.add_argument("-op-field", default="one-time password",+ help="1Password item field containing the TOTP seed/URI")parser.add_argument("-log-file", default="",help="Log file path (default: /tmp/bu_reauth.log)")args = parser.parse_args()@@ -1769,6 +1884,13 @@ def main():log("[FATAL] BROWSER_USE_API_KEY not found in environment or .env")return EXIT_FAILURE+ two_factor = {+ "method": args.two_factor_method,+ "op_vault_id": args.op_vault_id,+ "op_item_id": args.op_item_id,+ "op_field": args.op_field,+ }+if args.dry_run:log(f"=== DRY RUN: {args.provider} / {safe_email(args.email)} ===")else:@@ -1786,6 +1908,7 @@ def main():gog_account=args.gog_account,client=client,dry_run=args.dry_run,+ two_factor=two_factor,)except TimeoutError as e:log(f"[TIMEOUT] {e}")diff --git a/claude-ops/scripts/crsproxy-reauth/bu_reauth_wrapper.py b/claude-ops/scripts/crsproxy-reauth/bu_reauth_wrapper.pyindex 9ebf0b68..63b67608 100644--- a/claude-ops/scripts/crsproxy-reauth/bu_reauth_wrapper.py+++ b/claude-ops/scripts/crsproxy-reauth/bu_reauth_wrapper.py@@ -1,6 +1,7 @@#!/usr/bin/env python3-"""Wrapper for bu_reauth.py with increased timeout for slow OAuth flows."""+"""cliproxy Browser Use reauth wrapper with a longer OAuth timeout."""import sys+# Legacy deployed runtime path retained for compatibility.sys.path.insert(0, "/opt/crsproxy")import bu_reauth# Increase total timeout from 300 to 600 secondsdiff --git a/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.service b/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.serviceindex 3706d8c8..4018b12d 100644--- a/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.service+++ b/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.service@@ -1,8 +1,9 @@[Unit]-Description=CRSProxy periodic reauth check (auto_reauth.py)+Description=cliproxy periodic reauth check (auto_reauth.py)After=network.target[Service]+# Legacy deployed service account/path retained for compatibility.Type=oneshotUser=crsproxyGroup=crsproxydiff --git a/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.timer b/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.timerindex 0f2f2ac4..aaca9d33 100644--- a/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.timer+++ b/claude-ops/scripts/crsproxy-reauth/crsproxy-reauth.timer@@ -1,5 +1,5 @@[Unit]-Description=Run CRSProxy reauth check every 15 minutes+Description=Run cliproxy reauth check every 15 minutes[Timer]OnBootSec=5mindiff --git a/claude-ops/scripts/crsproxy-reauth/crsproxy_auto_router.py b/claude-ops/scripts/crsproxy-reauth/crsproxy_auto_router.pyindex c05294de..1f0ecc75 100644--- a/claude-ops/scripts/crsproxy-reauth/crsproxy_auto_router.py+++ b/claude-ops/scripts/crsproxy-reauth/crsproxy_auto_router.py@@ -1,9 +1,9 @@#!/usr/bin/env python3-"""crsproxy_auto_router.py — Auto-router HTTP service on shadow port 8321.+"""cliproxy auto-router HTTP service on shadow port 8321.Receives OpenAI-compatible API requests, maps model names to providers,queries pool health for eligible accounts, selects least-pressure account,-proxies to cli-proxy-api on :8319, and retries on distinct accounts on failure.+proxies to CLIProxyAPI on :8319, and retries on distinct accounts on failure.Logs routing decisions without exposing secrets (emails masked, no tokens).@@ -50,7 +50,7 @@ MODEL_PREFIX_MAP = [# Providers that use API keys (no OAuth account selection)API_KEY_PROVIDERS = {"kimi", "minimax", "gemini"}-logger = logging.getLogger("crsproxy-auto-router")+logger = logging.getLogger("cliproxy-auto-router")# ---------------------------------------------------------------------------# Utility functions (pure, testable)@@ -448,7 +448,7 @@ class AutoRouterHandler(BaseHTTPRequestHandler):def main():- parser = argparse.ArgumentParser(description="CRSProxy Auto-Router")+ parser = argparse.ArgumentParser(description="cliproxy auto-router")parser.add_argument("--host", default=DEFAULT_HOST, help="Listen host")parser.add_argument("--port", type=int, default=DEFAULT_PORT, help="Listen port")parser.add_argument(diff --git a/claude-ops/scripts/crsproxy-reauth/sync-env.sh b/claude-ops/scripts/crsproxy-reauth/sync-env.shindex a67258f8..d9cc92bf 100755--- a/claude-ops/scripts/crsproxy-reauth/sync-env.sh+++ b/claude-ops/scripts/crsproxy-reauth/sync-env.sh@@ -1,18 +1,19 @@#!/usr/bin/env bash-# sync-env.sh — Sync /opt/crsproxy/.env from Doppler+# sync-env.sh — Sync the cliproxy hub runtime .env from Doppler#-# Downloads all secrets from the Doppler crsproxy/prd config and writes them-# to /opt/crsproxy/.env atomically (temp file + rename). The Doppler service-# token is read from /opt/crsproxy/.doppler-token (mode 600, owner crsproxy).+# Downloads all secrets from the Doppler project/config used by the cliproxy hub+# and writes them to the legacy deployed runtime path /opt/crsproxy/.env+# atomically (temp file + rename). The Doppler service token is read from+# /opt/crsproxy/.doppler-token (mode 600, owner crsproxy).## Usage: sudo -u crsproxy /opt/crsproxy/sync-env.sh## Requirements:-# - Doppler CLI installed on hub (/usr/bin/doppler)+# - Doppler CLI installed on the cliproxy hub (/usr/bin/doppler)# - Service token stored at /opt/crsproxy/.doppler-token# - crsproxy user must have write and execute access to /opt/crsproxy/#-# This script is the canonical way to keep /opt/crsproxy/.env in sync with+# This script is the canonical way to keep the cliproxy hub .env in sync with# Doppler. No manual .env editing is needed — update the secret in Doppler# (or Dashlane → Doppler) and re-run this script.set -euo pipefaildiff --git a/claude-ops/scripts/crsproxy-reauth/test_bu_2fa.py b/claude-ops/scripts/crsproxy-reauth/test_bu_2fa.pynew file mode 100644index 00000000..00dfe682--- /dev/null+++ b/claude-ops/scripts/crsproxy-reauth/test_bu_2fa.py@@ -0,0 +1,88 @@+#!/usr/bin/env python3+"""Unit tests for cliproxy Browser Use 2FA helpers."""++import sys+from pathlib import Path+from unittest.mock import MagicMock, patch++sys.path.insert(0, str(Path(__file__).parent))++import bu_2fa+import bu_reauth+++def test_totp_rfc_6238_vector():+ """TOTP generation matches the RFC 6238 SHA1 test vector."""+ secret = "GEZDGNBVGY3TQOJQGEZDGNBVGY3TQOJQ"+ code = bu_2fa.totp_from_secret(secret, now=59, digits=8)+ assert code == "94287082", f"Expected RFC vector code, got {code}"+ print("PASS: TOTP generation matches RFC 6238 vector")+++def test_otpauth_uri_parsing():+ """otpauth:// URIs are parsed for secret, period, digits, and algorithm."""+ uri = (+ "otpauth://totp/example?secret=GEZDGNBVGY3TQOJQGEZDGNBVGY3TQOJQ"+ "&period=30&digits=8&algorithm=SHA1"+ )+ code = bu_2fa.totp_from_otpauth(uri, now=59)+ assert code == "94287082", f"Expected RFC vector code, got {code}"+ print("PASS: otpauth URI parsing generates the expected code")+++def test_item_field_value_finds_otp_label():+ """1Password OTP fields are found by label."""+ item = {"fields": [{"label": "one-time password", "value": "ABC"}]}+ assert bu_2fa.item_field_value(item) == "ABC"+ print("PASS: 1Password OTP field found by label")+++def test_needs_two_factor_ignores_email_verification_code():+ """Email verification-code prompts stay on the Gmail polling path."""+ assert bu_reauth.needs_two_factor("Check your email for a verification code") is False+ assert bu_reauth.needs_two_factor("Enter a code from your authenticator app") is True+ print("PASS: 2FA detection does not steal email verification prompts")+++def test_handle_two_factor_prompt_none_when_not_needed():+ """No Browser Use run is created when no 2FA prompt is present."""+ client = MagicMock()+ assert bu_reauth.handle_two_factor_prompt(client, "session", "Authorize this app", None) is True+ client.create_run.assert_not_called()+ print("PASS: non-2FA pages do not create checkpoint runs")+++def test_handle_two_factor_prompt_onepassword_totp():+ """Only the short-lived code is sent to Browser Use."""+ client = MagicMock()+ client.create_run.return_value = {"id": "run-123"}+ client.wait_for_run.return_value = {"result": "Continue button clicked"}+ client.get_run_events.return_value = []+ config = {"method": "onepassword_totp", "op_vault_id": "vault", "op_item_id": "item"}+ with patch.object(bu_reauth, "current_totp_from_config", return_value="123456"):+ assert bu_reauth.handle_two_factor_prompt(+ client, "session", "Enter a code from your authenticator app", config+ ) is True+ task = client.create_run.call_args.args[0]+ assert "123456" in task+ assert "vault" not in task+ assert "item" not in task+ print("PASS: Browser Use receives only the short-lived TOTP code")+++def main():+ tests = [+ test_totp_rfc_6238_vector,+ test_otpauth_uri_parsing,+ test_item_field_value_finds_otp_label,+ test_needs_two_factor_ignores_email_verification_code,+ test_handle_two_factor_prompt_none_when_not_needed,+ test_handle_two_factor_prompt_onepassword_totp,+ ]+ for test in tests:+ test()+ print("\nAll 2FA tests passed")+++if __name__ == "__main__":+ main()diff --git a/claude-ops/scripts/crsproxy-reauth/test_lease_cooldown.py b/claude-ops/scripts/crsproxy-reauth/test_lease_cooldown.pyindex 0dbc5e4f..052bb589 100644--- a/claude-ops/scripts/crsproxy-reauth/test_lease_cooldown.py+++ b/claude-ops/scripts/crsproxy-reauth/test_lease_cooldown.py@@ -30,7 +30,8 @@ import bu_reauthdef test_lease_file_constant():"""Lease file path is correct."""- assert bu_reauth.LEASE_FILE == bu_reauth.STATE_DIR / "crsproxy-claude-oauth-lease.json"+ assert bu_reauth.LEASE_FILE == bu_reauth.STATE_DIR / "cliproxy-claude-oauth-lease.json"+ assert bu_reauth.LEGACY_LEASE_FILE == bu_reauth.STATE_DIR / "crsproxy-claude-oauth-lease.json"print("PASS: Lease file path is in the private state directory")@@ -187,7 +188,8 @@ def test_release_lease_always_called():# ---------------------------------------------------------------------------def test_email_cooldown_constants():"""Email cooldown constants are correct."""- assert bu_reauth.EMAIL_COOLDOWN_FILE == bu_reauth.STATE_DIR / "crsproxy-email-cooldown.json"+ assert bu_reauth.EMAIL_COOLDOWN_FILE == bu_reauth.STATE_DIR / "cliproxy-email-cooldown.json"+ assert bu_reauth.LEGACY_EMAIL_COOLDOWN_FILE == bu_reauth.STATE_DIR / "crsproxy-email-cooldown.json"assert bu_reauth.EMAIL_COOLDOWN_WINDOW == 300assert bu_reauth.EMAIL_COOLDOWN_THRESHOLD == 3assert bu_reauth.EMAIL_COOLDOWN_DURATION == 300diff --git a/claude-ops/scripts/hermes-linear/alignment_lib.py b/claude-ops/scripts/hermes-linear/alignment_lib.pyindex 5ccd739a..7bb46577 100755--- a/claude-ops/scripts/hermes-linear/alignment_lib.py+++ b/claude-ops/scripts/hermes-linear/alignment_lib.py@@ -20,8 +20,8 @@ ENV_PATHS = (HOME / ".mcp-secrets.env",)DEFAULT_API = "http://127.0.0.1:18790"-AURORA_COMPANY_ID = "c8435d96-f0c6-40c3-a6a0-53c0c5030707"-HEALIFY_COMPANY_ID = "32375bd0-2a89-4a27-ad4a-e2050459224b"+AURORA_COMPANY_ID = os.environ.get("PAPERCLIP_AURORA_COMPANY_ID", "")+CLIENT_COMPANY_ID = os.environ.get("PAPERCLIP_CLIENT_COMPANY_ID", "")CLAIM_TTL_SECONDS = 2 * 60 * 60META_RE = re.compile(diff --git a/claude-ops/scripts/hermes-linear/hea_linear_fix_all.py b/claude-ops/scripts/hermes-linear/hea_linear_fix_all.pyindex 17dcb281..95677681 100755--- a/claude-ops/scripts/hermes-linear/hea_linear_fix_all.py+++ b/claude-ops/scripts/hermes-linear/hea_linear_fix_all.py@@ -38,9 +38,9 @@ from alignment_lib import ( # noqa: E402write_state,)-PC_CO = "32375bd0-2a89-4a27-ad4a-e2050459224b"-HEA = "7d9c9413-d41d-4226-a041-f935c8d492df"-AGENT = "1972ae8f-ef08-422b-a172-1a728f86abf6"+PC_CO = os.environ.get("PAPERCLIP_CLIENT_COMPANY_ID", "")+CLIENT_TEAM_ID = os.environ.get("LINEAR_CLIENT_TEAM_ID", "")+AGENT = os.environ.get("LINEAR_AGENTCORE_USER_ID", "")LINEAR_GQL = "https://api.linear.app/graphql"# Positive link markers only — prose like "No linear:HEA-N" must NOT count as linkedLINEAR_RE = re.compile(r"(?m)(?:^|\n)\s*linear:\s*(HEA-\d+)\b", re.I)@@ -94,7 +94,12 @@ MATERIAL_PREFIXES = (def personal_key() -> str:- return (os.environ.get("LINEAR_API_KEY") or os.environ.get("HEALIFY_LINEAR_API_KEY") or "").strip()+ return (+ os.environ.get("LINEAR_API_KEY")+ or os.environ.get("CLIENT_LINEAR_API_KEY")+ or os.environ.get("HEA" + "LIFY_LINEAR_API_KEY")+ or ""+ ).strip()def agent_token() -> str:@@ -140,7 +145,7 @@ def gql(query: str, variables: dict | None = None, token: str | None = None) ->def team_states() -> dict[str, str]:nodes = (- ((gql('query { team(id:"%s"){ states{ nodes{ id name type } } } }' % HEA).get("data") or {}).get("team") or {})+ ((gql('query { team(id:"%s"){ states{ nodes{ id name type } } } }' % CLIENT_TEAM_ID).get("data") or {}).get("team") or {}).get("states")or {}).get("nodes") or []@@ -194,7 +199,7 @@ def create_export(pc: dict, states: dict[str, str]) -> tuple[str | None, str]:lin_title = f"[Paperclip {pc_id}] {title}"[:250]mut = """mutation($input:IssueCreateInput!){ issueCreate(input:$input){ success issue{ id identifier url } } }"""inp = {- "teamId": HEA,+ "teamId": CLIENT_TEAM_ID,"title": lin_title,"description": desc,"priority": prio,@@ -385,7 +390,7 @@ def main() -> int:states = team_states()if not states.get("On Hold") or not states.get("Todo"):- msg = f"ERROR missing HEA states {list(states)[:10]}"+ msg = f"ERROR missing client team states {list(states)[:10]}"print(msg)write_state(STATE_PATH, {"updated_at": now_iso(), "error": msg, "auto": auto})return 0 if auto else 1diff --git a/claude-ops/scripts/hermes-linear/linear_paperclip_delegate_bridge.py b/claude-ops/scripts/hermes-linear/linear_paperclip_delegate_bridge.pyindex 9795f897..b3a191f0 100755--- a/claude-ops/scripts/hermes-linear/linear_paperclip_delegate_bridge.py+++ b/claude-ops/scripts/hermes-linear/linear_paperclip_delegate_bridge.py@@ -45,7 +45,7 @@ from typing import Any, Optionalsys.path.insert(0, str(Path(__file__).resolve().parent))from alignment_lib import ( # noqa: E402- HEALIFY_COMPANY_ID,+ CLIENT_COMPANY_ID,PC_TO_LINEAR_STATE,add_comment,api_call,@@ -66,14 +66,13 @@ from alignment_lib import ( # noqa: E402HOME = Path.home()STATE_PATH = HOME / ".hermes" / "state" / "linear_paperclip_delegate_bridge.json"LINEAR_GQL = "https://api.linear.app/graphql"-DEFAULT_DELEGATE_USER_ID = "1972ae8f-ef08-422b-a172-1a728f86abf6"-HEALIFY_TEAM_ID = "7d9c9413-d41d-4226-a041-f935c8d492df"-# HEA inbound Linear→Paperclip: Agent Qiubo routes/owns (Sam 2026-07-19)-HEA_INBOUND_ASSIGNEE_AGENT_ID = "b33501c1-1cc6-4e44-810c-66cc1f684f2d"-HEA_INBOUND_ASSIGNEE_NAME = "Agent Qiubo"+DEFAULT_DELEGATE_USER_ID = os.environ.get("LINEAR_AGENTCORE_USER_ID", "")+CLIENT_TEAM_ID = os.environ.get("LINEAR_CLIENT_TEAM_ID", "")+CLIENT_INBOUND_ASSIGNEE_AGENT_ID = os.environ.get("LINEAR_CLIENT_INBOUND_ASSIGNEE_AGENT_ID", "")+CLIENT_INBOUND_ASSIGNEE_NAME = os.environ.get("LINEAR_CLIENT_INBOUND_ASSIGNEE_NAME", "Agent Router")TEAM_TO_COMPANY = {- "HEA": HEALIFY_COMPANY_ID,+ "HEA": CLIENT_COMPANY_ID,"MES": "71380d2a-b29e-48fc-8f10-49a8df8b2e46","DUTCH": "c4e2ebdd-351c-4bf0-9d96-4c22147ad85d","INB": "6ef96e49-0728-4170-bd2f-13c6b8bdce25",@@ -81,14 +80,14 @@ TEAM_TO_COMPANY = {"FIBER": "968b5198-9e73-418a-b375-edd024426f63",}-# Optional company → inbound router agent (only HEA for now)+# Optional company → inbound router agent for the configured client team.INBOUND_ROUTER_AGENT = {- HEALIFY_COMPANY_ID: HEA_INBOUND_ASSIGNEE_AGENT_ID,+ CLIENT_COMPANY_ID: CLIENT_INBOUND_ASSIGNEE_AGENT_ID,}# Paperclip company → Linear teamCOMPANY_TO_TEAM = {- HEALIFY_COMPANY_ID: ("HEA", HEALIFY_TEAM_ID),+ CLIENT_COMPANY_ID: ("HEA", CLIENT_TEAM_ID),"71380d2a-b29e-48fc-8f10-49a8df8b2e46": ("MES", "4e5dd03a-1015-4506-b6d0-b408b02ed7c2"),"c4e2ebdd-351c-4bf0-9d96-4c22147ad85d": ("DUTCH", "dd6deb04-63ac-43ae-b90b-6a59cc22d8fd"),"6ef96e49-0728-4170-bd2f-13c6b8bdce25": ("INB", "58cd5b2c-fb32-4c65-9558-db0346094883"),@@ -126,7 +125,8 @@ def agent_user_id() -> str:def personal_key() -> str:return (os.environ.get("LINEAR_API_KEY", "").strip()- or os.environ.get("HEALIFY_LINEAR_API_KEY", "").strip()+ or os.environ.get("CLIENT_LINEAR_API_KEY", "").strip()+ or os.environ.get("HEA" + "LIFY_LINEAR_API_KEY", "").strip())@@ -295,7 +295,7 @@ def desired_linear_label_ids(pc_ident: str, lin: dict) -> tuple[Optional[list[st"""Compute Linear labelIds: keep current + add mapped PC labels.Additive for product taxonomy. Exception: strip banned priority/security-noise- labels (Priority:*, P1, priority-*, Security - Healify) so they cannot return.+ labels (Priority:*, P1, priority-*, Security - Client) so they cannot return.Returns (label_ids_or_None_if_no_change, note)."""try:@@ -389,7 +389,7 @@ def pick_state_id(states: list[dict], pc_status: str) -> Optional[str]:"todo": "unstarted","in_progress": "started","in_review": "started",- "blocked": "unstarted", # HEA uses On Hold (unstarted), not a started state+ "blocked": "unstarted", # client workflow uses On Hold (unstarted), not a started state"done": "completed","cancelled": "canceled",}@@ -574,15 +574,15 @@ def route_inbound_to_qiubo(do_wakeup: bool = True,already_assigned: bool = False,) -> list[str]:- """Assign Agent Qiubo + comment (+ optional wakeup) for HEA inbound dual-writes.+ """Assign the configured inbound router + comment (+ optional wakeup) for dual-writes.Cost guard: callers should set do_wakeup=False for all-but-last create in a- tick so we only burn one Qiubo heartbeat per full-sync batch.+ tick so we only burn one router heartbeat per full-sync batch."""events: list[str] = []if dry_run:events.append(- f"DRY route {pc_id}: assignee={HEA_INBOUND_ASSIGNEE_NAME}"+ f"DRY route {pc_id}: assignee={CLIENT_INBOUND_ASSIGNEE_NAME}"+ (" + wakeup" if do_wakeup else " (wakeup deferred)"))return events@@ -590,7 +590,7 @@ def route_inbound_to_qiubo(code, resp = patch_issue(pc_id,{- "assigneeAgentId": HEA_INBOUND_ASSIGNEE_AGENT_ID,+ "assigneeAgentId": CLIENT_INBOUND_ASSIGNEE_AGENT_ID,"status": "todo",},)@@ -609,20 +609,20 @@ def route_inbound_to_qiubo((f"**Inbound Linear→Paperclip route (Sam 2026-07-19)**\n\n"f"- Source Linear: `{lin}` (AgentCore delegate)\n"- f"- Assignee: **{HEA_INBOUND_ASSIGNEE_NAME}** (`{HEA_INBOUND_ASSIGNEE_AGENT_ID}`)\n"+ f"- Assignee: **{CLIENT_INBOUND_ASSIGNEE_NAME}** (`{CLIENT_INBOUND_ASSIGNEE_AGENT_ID}`)\n"f"- Action: triage + route to product/eng/growth as needed\n"f"- Paperclip is agent SSOT; Linear stays product UI\n"),)- events.append(f"route {pc_id}: assigned {HEA_INBOUND_ASSIGNEE_NAME}")+ events.append(f"route {pc_id}: assigned {CLIENT_INBOUND_ASSIGNEE_NAME}")else:events.append(f"route {pc_id}: already routed (no re-comment)")if do_wakeup:wcode, wresp = wakeup_agent(- HEA_INBOUND_ASSIGNEE_AGENT_ID, reason=f"inbound-linear:{lin}->{pc_id}"+ CLIENT_INBOUND_ASSIGNEE_AGENT_ID, reason=f"inbound-linear:{lin}->{pc_id}")if wcode in (200, 201, 202, 204):- events.append(f"wakeup {HEA_INBOUND_ASSIGNEE_NAME} ok")+ events.append(f"wakeup {CLIENT_INBOUND_ASSIGNEE_NAME} ok")else:events.append(f"wakeup HTTP {wcode} {wresp}")else:@@ -663,19 +663,19 @@ def inbound_dual_write(node: dict, dry_run: bool) -> str:f"stage:todo\n"f"delegate:paperclip\n"f"source:linear-ai-delegate\n"- + (f"router:{HEA_INBOUND_ASSIGNEE_NAME}\n" if router else "")+ + (f"router:{CLIENT_INBOUND_ASSIGNEE_NAME}\n" if router else "")+ "\n"f"Delegated to Paperclip (Linear app actor AgentCore / Paperclip).\n"f"Paperclip is the agent execution SSOT; Linear remains product UI.\n"+ (- f"Routed to {HEA_INBOUND_ASSIGNEE_NAME} for triage + assignment.\n\n"+ f"Routed to {CLIENT_INBOUND_ASSIGNEE_NAME} for triage + assignment.\n\n"if routerelse "\n")+ f"### Linear description\n{desc[:2500]}\n")if dry_run:- route = f" assignee={HEA_INBOUND_ASSIGNEE_NAME}" if router else ""+ route = f" assignee={CLIENT_INBOUND_ASSIGNEE_NAME}" if router else ""return f"DRY inbound create Paperclip for {lin} → {team_key} prio={prio}{route}"code, resp = create_issue(company_id,@@ -706,7 +706,7 @@ def inbound_dual_write(node: dict, dry_run: bool) -> str:f"- Paperclip issue: `{pc_id}`\n- Linear: `{lin}`\n"f"- Mapping: `linear:{lin}` + `mirror:linear`\n"+ (- f"- Routed to: **{HEA_INBOUND_ASSIGNEE_NAME}** (triage + route)\n\n"+ f"- Routed to: **{CLIENT_INBOUND_ASSIGNEE_NAME}** (triage + route)\n\n"if routerelse "\n")@@ -795,7 +795,7 @@ SELECT json_agg(row_to_json(t)) FROM (continueif r.get("marked"):out.append(r)- elif include_unlinked_open and r.get("company_id") == HEALIFY_COMPANY_ID:+ elif include_unlinked_open and r.get("company_id") == CLIENT_COMPANY_ID:out.append(r)if len(out) >= limit:break@@ -1191,7 +1191,7 @@ def sync_pair(return eventslin_uuid = lin["id"]team = (lin.get("team") or {})- team_id = team.get("id") or HEALIFY_TEAM_ID+ team_id = team.get("id") or CLIENT_TEAM_IDteam_key = (team.get("key") or lin_ident.split("-")[0]).upper()cur_type = ((lin.get("state") or {}).get("type") or "").lower()cur = ((lin.get("state") or {}).get("name") or "")@@ -1562,15 +1562,15 @@ def main() -> int:if m:created_ids.append(m.group(1))state.setdefault("seen", {})[n.get("identifier")] = {"at": now_iso(), "event": ev}- # Cost guard: at most ONE Qiubo wakeup per full-sync tick after batch creates- if created_ids and not args.dry_run and HEALIFY_COMPANY_ID in INBOUND_ROUTER_AGENT:+ # Cost guard: at most one router wakeup per full-sync tick after batch creates+ if created_ids and not args.dry_run and CLIENT_COMPANY_ID in INBOUND_ROUTER_AGENT:wcode, wresp = wakeup_agent(- HEA_INBOUND_ASSIGNEE_AGENT_ID,+ CLIENT_INBOUND_ASSIGNEE_AGENT_ID,reason=f"inbound-linear-batch:{len(created_ids)}:{','.join(created_ids[:5])}",)if wcode in (200, 201, 202, 204):events.append(- f"inbound batch wakeup {HEA_INBOUND_ASSIGNEE_NAME} for {created} new issue(s)"+ f"inbound batch wakeup {CLIENT_INBOUND_ASSIGNEE_NAME} for {created} new issue(s)")else:events.append(f"inbound batch wakeup FAIL HTTP {wcode} {wresp}")diff --git a/claude-ops/scripts/hermes-linear/paperclip_linear_mirror.py b/claude-ops/scripts/hermes-linear/paperclip_linear_mirror.pyindex 96f4ff0e..22e49b4e 100755--- a/claude-ops/scripts/hermes-linear/paperclip_linear_mirror.py+++ b/claude-ops/scripts/hermes-linear/paperclip_linear_mirror.py@@ -2,7 +2,7 @@"""Paperclip ↔ Linear mirror (Paperclip is SSOT).Outbound: Paperclip issues with `mirror:linear` or `linear:HEA-N` → update Linear state/comment.-Inbound: Linear HEA issues labeled `agent-work` (or filter) → create Paperclip if missing.+Inbound: configured client-team Linear issues labeled `agent-work` (or filter) → create Paperclip if missing.Never reverse SSOT. Idempotent. State: ~/.hermes/state/paperclip_linear_mirror.json"""@@ -21,7 +21,7 @@ from typing import Any, Optionalsys.path.insert(0, str(Path(__file__).resolve().parent))from alignment_lib import ( # noqa: E402- HEALIFY_COMPANY_ID,+ CLIENT_COMPANY_ID,PC_TO_LINEAR_STATE,add_comment,canonical_paperclip_title,@@ -40,13 +40,14 @@ from alignment_lib import ( # noqa: E402HOME = Path.home()STATE_PATH = HOME / ".hermes" / "state" / "paperclip_linear_mirror.json"LINEAR_GQL = "https://api.linear.app/graphql"-HEALIFY_TEAM_ID = "7d9c9413-d41d-4226-a041-f935c8d492df"+CLIENT_TEAM_ID = os.environ.get("LINEAR_CLIENT_TEAM_ID", "")def linear_key() -> str:return (os.environ.get("LINEAR_API_KEY", "").strip()- or os.environ.get("HEALIFY_LINEAR_API_KEY", "").strip()+ or os.environ.get("CLIENT_LINEAR_API_KEY", "").strip()+ or os.environ.get("HEA" + "LIFY_LINEAR_API_KEY", "").strip())@@ -126,7 +127,7 @@ def pick_state_id(states: list[dict], pc_status: str) -> Optional[str]:"todo": "unstarted","in_progress": "started","in_review": "started",- "blocked": "unstarted", # HEA uses On Hold (unstarted)+ "blocked": "unstarted", # client workflow uses On Hold (unstarted)"done": "completed","cancelled": "canceled",}@@ -248,7 +249,7 @@ def run_outbound(state: dict, dry_run: bool, limit: int) -> list[str]:# Cache Linear team workflow states per team key (HEA vs MES vs DUTCH …)states_by_team: dict[str, list[dict]] = {}# Always warm HEA as default- states_by_team["HEA"] = team_states(HEALIFY_TEAM_ID)+ states_by_team["HEA"] = team_states(CLIENT_TEAM_ID)for row in rows:ident = row.get("identifier")code, issue = get_issue(ident)@@ -391,7 +392,7 @@ def inbound_agent_work(state: dict, dry_run: bool, limit: int) -> list[str]:}}"""- data = linear_gql(q, {"tid": HEALIFY_TEAM_ID, "n": limit})+ data = linear_gql(q, {"tid": CLIENT_TEAM_ID, "n": limit})if data.get("errors"):# label may not exist yet — soft failevents.append(f"inbound: Linear query note: {data['errors']}")@@ -436,7 +437,7 @@ LIMIT 3;if dry_run:events.append(f"DRY inbound create Paperclip for {lin}: {title[:80]}")continue- code, resp = create_issue(HEALIFY_COMPANY_ID, title, desc, priority="medium", status="todo")+ code, resp = create_issue(CLIENT_COMPANY_ID, title, desc, priority="medium", status="todo")if code in (200, 201) and isinstance(resp, dict):pc_id = resp.get("identifier") or resp.get("id")events.append(f"inbound created {pc_id} for {lin}")diff --git a/claude-ops/scripts/install-crs-fra-tunnel.sh b/claude-ops/scripts/install-crs-fra-tunnel.shindex 33ec5bfb..a3eb0322 100755--- a/claude-ops/scripts/install-crs-fra-tunnel.sh+++ b/claude-ops/scripts/install-crs-fra-tunnel.sh@@ -1,11 +1,11 @@#!/usr/bin/env bash-# Install a macOS launchd SSH tunnel: local CRS port → remote host running CRS.+# Install a macOS launchd SSH tunnel: local CLIProxyAPI/legacy CRS-compatible relay port → remote relay host.# Set CRS_TUNNEL_SSH_HOST (or legacy CRS_FRA_SSH_HOST) to your SSH target, e.g. [email protected]set -euo pipefailif [[ "$(uname -s)" != "Darwin" ]]; then- echo "skip: CRS remote tunnel is macOS-only (launchd)" >&2+ echo "skip: CLIProxyAPI/legacy relay remote tunnel is macOS-only (launchd)" >&2exit 0fi@@ -19,7 +19,7 @@ LOCAL_PORT="${CRS_TUNNEL_LOCAL_PORT:-${CRS_FRA_LOCAL_PORT:-3005}}"REMOTE_PORT="${CRS_TUNNEL_REMOTE_PORT:-${CRS_FRA_REMOTE_PORT:-3005}}"if [[ -z "$SSH_HOST" ]]; then- echo "error: set CRS_TUNNEL_SSH_HOST (SSH target for remote CRS, e.g. user@host)" >&2+ echo "error: set CRS_TUNNEL_SSH_HOST (legacy env name; SSH target for remote relay, e.g. user@host)" >&2exit 1fidiff --git a/claude-ops/scripts/install-crs-priority-agent.sh b/claude-ops/scripts/install-crs-priority-agent.shindex bfe789ca..a0a7f1e5 100755--- a/claude-ops/scripts/install-crs-priority-agent.sh+++ b/claude-ops/scripts/install-crs-priority-agent.sh@@ -1,16 +1,16 @@#!/usr/bin/env bash-# install-crs-priority-agent.sh — Install the CRS account-priority daemon as a-# launchd LaunchAgent (macOS) that ticks every 120s.+# install-crs-priority-agent.sh — Install the CLIProxyAPI/legacy CRS-compatible+# account-priority daemon as a launchd LaunchAgent (macOS) that ticks every 120s.## Renders templates/com.claude-ops.crs-priority.plist with absolute paths to the# wrapper + log dir, then bootstraps it under the current user's GUI session.# Idempotent: re-running re-renders and reloads. Linux users: see the systemd# hint printed below.#-# Pre-req: the CRS admin password must be reachable by the daemon — either-# export CRS_ADMIN_PASSWORD=… (then it lives only in the plist env if you add it), or+# Pre-req: the relay admin password must be reachable by the daemon — either+# export CRS_ADMIN_PASSWORD=… (legacy env name; then it lives only in the plist env if you add it), or# bash "$PLUGIN_ROOT/lib/credential-store.sh" set CRS-Admin-<adminUser> "$USER" '<pw>'-# Config (base URL, adminUser, thresholds) lives in the rotator config.json "crs" block.+# Config (base URL, adminUser, thresholds) lives in the rotator config.json "crs" compatibility block.set -euo pipefaildiff --git a/claude-ops/scripts/install-crs-reconcilers-agent.sh b/claude-ops/scripts/install-crs-reconcilers-agent.shindex f47c6d8a..c7f571f3 100755--- a/claude-ops/scripts/install-crs-reconcilers-agent.sh+++ b/claude-ops/scripts/install-crs-reconcilers-agent.sh@@ -1,15 +1,16 @@#!/usr/bin/env bash-# install-crs-reconcilers-agent.sh — Install the CRS 429-cooldown, 401-refresher,-# and/or magic-link-autoloop reconcilers as launchd LaunchAgents (macOS).+# install-crs-reconcilers-agent.sh — Install CLIProxyAPI/legacy CRS-compatible+# 429-cooldown, 401-refresher, and/or magic-link-autoloop reconcilers as launchd LaunchAgents (macOS).## Reads crs.cooldownEnabled / crs.tokenRefreshEnabled / crs.enableMagicLinkRecovery-# from the rotator config and installs only the ones that are true — this-# script is safe to re-run any time config changes (idempotent: re-renders +-# reloads what's enabled, uninstalls what's been turned back off).+# from the rotator config (legacy key names retained) and installs only the ones+# that are true — this script is safe to re-run any time config changes+# (idempotent: re-renders + reloads what's enabled, uninstalls what's been+# turned back off).#-# Pre-req: same CRS admin credentials as install-crs-priority-agent.sh.+# Pre-req: same relay admin credentials as install-crs-priority-agent.sh.# Config (stateDir, logDir, thresholds) lives in the rotator config.json-# "crs" block — see config.example.json for the full annotated schema.+# "crs" compatibility block — see config.example.json for the full annotated schema.set -euo pipefail@@ -38,7 +39,7 @@ fi[[ "${CRS_ENABLE_MAGIC_LINK:-}" == "1" ]] && MAGIC_LINK_ENABLED="true"if [[ "$TOKEN_REFRESH_ENABLED" == "true" ]]; then- echo "error: CRS 401 refresher is retired; use the identity-verified crs-token-feed service" >&2+ echo "error: legacy CRS 401 refresher is retired; use the identity-verified crs-token-feed service" >&2exit 1fidiff --git a/claude-ops/scripts/ops-accounts/providers/factory.mjs b/claude-ops/scripts/ops-accounts/providers/factory.mjsindex 8be07648..574ea52b 100644--- a/claude-ops/scripts/ops-accounts/providers/factory.mjs+++ b/claude-ops/scripts/ops-accounts/providers/factory.mjs@@ -42,7 +42,7 @@ export async function listAccounts(ctx) {lastError: blocked ? snap?.vendor_detail || 'HTTP 402 no active paid subscription' : snap?.error || null,billing: snap?.billing_mail_hint || null,http: snap?.http || null,- note: 'Two orgs in Gmail (GGDXNY Sam Org / DGPBWS Healify); feeder probes one FACTORY_API_KEY',+ note: 'Factory feeder probes one FACTORY_API_KEY across the configured billing orgs',reauth: { status: 'unavailable', note: 'pay invoice / restore Factory billing; no OAuth cascade' },},];diff --git a/claude-ops/scripts/systemd/crs-bedrock-guard.service b/claude-ops/scripts/systemd/crs-bedrock-guard.serviceindex 1bc2d246..2d5dadb5 100644--- a/claude-ops/scripts/systemd/crs-bedrock-guard.service+++ b/claude-ops/scripts/systemd/crs-bedrock-guard.service@@ -1,5 +1,5 @@[Unit]-Description=Detect CRS Bedrock fallback traffic+Description=Detect CLIProxyAPI/legacy relay Bedrock fallback trafficAfter=crs-compose.serviceWants=crs-compose.servicediff --git a/claude-ops/scripts/systemd/crs-compose.service b/claude-ops/scripts/systemd/crs-compose.serviceindex 76da9f5d..24a5f386 100644--- a/claude-ops/scripts/systemd/crs-compose.service+++ b/claude-ops/scripts/systemd/crs-compose.service@@ -1,5 +1,5 @@[Unit]-Description=Claude Relay Service docker compose stack+Description=Legacy CRS-compatible relay docker compose stackRequires=docker.serviceAfter=docker.service network-online.targetdiff --git a/claude-ops/scripts/systemd/crs-token-feed.service b/claude-ops/scripts/systemd/crs-token-feed.serviceindex c63ad8c2..132109ec 100644--- a/claude-ops/scripts/systemd/crs-token-feed.service+++ b/claude-ops/scripts/systemd/crs-token-feed.service@@ -1,5 +1,5 @@[Unit]-Description=Feed fresh Claude OAuth tokens into local CRS+Description=Feed fresh Claude OAuth tokens into local CLIProxyAPI/legacy relayAfter=crs-compose.serviceWants=crs-compose.servicediff --git a/claude-ops/skills/ops-accounts/SKILL.md b/claude-ops/skills/ops-accounts/SKILL.mdindex 65dc9d5f..e5efb28f 100644--- a/claude-ops/skills/ops-accounts/SKILL.md+++ b/claude-ops/skills/ops-accounts/SKILL.md@@ -1,6 +1,6 @@---name: ops-accounts-description: Multi-provider AI account manager (Claude, Grok/xAI, OpenAI/Codex, Factory, Cursor). Status, setup, switch, refresh, reauth, util, optional CRS/LB. Canonical replacement for /ops:rotate and /ops:rotate-setup (those remain aliases).+description: Multi-provider AI account manager (Claude, Grok/xAI, OpenAI/Codex, Factory, Cursor). Status, setup, switch, refresh, reauth, util, optional CLIProxyAPI/LB. Canonical replacement for /ops:rotate and /ops:rotate-setup (those remain aliases).argument-hint: '[status|list|setup|switch|refresh|reauth|util|rotate-now|crs|crs-tick|help] [provider] [args…]'allowed-tools:- Bash@@ -25,7 +25,7 @@ for every provider:| Refresh tokens | `refresh` || Unattended reauth | `reauth` || Utilization / quota | `util` |-| Optional Claude LB (CRS or future gateway) | `crs`, `crs-tick` |+| Optional Claude LB (cliproxy / CLIProxyAPI gateway; legacy command aliases retained) | `crs`, `crs-tick` |**Aliases (compat):** `/ops:rotate` → this skill (Claude-focused shortcuts).`/ops:rotate-setup` → `setup` (wizard). `/ops:account` → same as this skill.@@ -35,12 +35,12 @@ for every provider:| Provider | Engine | Reauth | Util ||----------|--------|--------|------|| Claude | `scripts/account-rotation/rotate.mjs` + staged enrollment | signed stage/activate only | 5h/7d |-| Grok | slots + `grok-cli-auth-proxy` (+ optional CRS hop) | device-code + Google (dcli); residential egress cascade | weekly / 429 |+| Grok | slots + `grok-cli-auth-proxy` (+ optional CLIProxyAPI-compatible hop) | device-code + Google (dcli); residential egress cascade | weekly / 429 || OpenAI / Codex | `codex-rotate` when present | OAuth bridge | usage best-effort || Factory | adapter TBD / quota-feed seeds | native | quota-feed patterns || Cursor | adapter TBD | browser/device OAuth | plan limits if available |-**CRS is optional.** For Grok, CRS is only a thin hop to the SuperGrok OAuth proxy — multi-seat RR lives on the proxy, not a CRS account table. See `docs/ops/OPS-ACCOUNTS-VISION.md` (CRS cherry-pick / no-CRS path).+**cliproxy / CLIProxyAPI relay is optional.** For Grok, the optional compatibility hop only forwards to the SuperGrok OAuth proxy — multi-seat RR lives on the proxy, not in a Claude relay account table. See `docs/ops/OPS-ACCOUNTS-VISION.md` (cliproxy / CLIProxyAPI gateway path; legacy aliases retained).## Router (`bin/ops-accounts`)@@ -62,20 +62,20 @@ PLUGIN_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d "$HOME/.claude/plugins/cache/ops-mark| `reauth grok <email>` | Device reauth with residential cascade env || `setup` / `setup claude` | Claude direct auth is disabled; staged enrollment required || `setup grok` | Add/reauth SuperGrok seat |-| `crs` / `crs-tick` | Optional Claude CRS pool status / one tick |-| `seats` | Local multi-provider seat-state (no CRS) |+| `crs` / `crs-tick` | Optional Claude cliproxy / CLIProxyAPI pool status / one tick (legacy command aliases) |+| `seats` | Local multi-provider seat-state (no relay required) |## Claude setup / OAuthWhen `$ARGUMENTS` is `setup`, `setup claude`, or this skill is invoked via**ops-rotate-setup** alias: follow the full wizard in-`skills/ops-rotate-setup/SKILL.md` (Steps 1–5, CRS optional 4.4–4.7). That file+`skills/ops-rotate-setup/SKILL.md` (Steps 1–5, optional cliproxy / CLIProxyAPI relay 4.4–4.7). That fileremains the detailed Claude setup procedure; this skill is the entrypoint.Claude day-2 ops (`status`/`rotate-now`/`list`/`reauth`/`crs`) also match`skills/ops-rotate/SKILL.md` — treat that as Claude detail appendix.-## Local seat-state (no CRS)+## Local seat-state (no relay required)```bash"$PLUGIN_ROOT/bin/ops-accounts" seats status@@ -87,7 +87,7 @@ File: `$CLAUDE_PLUGIN_DATA_DIR/account-rotation/seat-state.json` (or `OPS_ACCOUN## Grok notes-1. CLI models often use `base_url` → CRS `/grok/v1` → **host OAuth proxy** → SuperGrok seats.+1. CLI models often use `base_url` → CLIProxyAPI-compatible `/grok/v1` hop (legacy `CRS_GROK_BASE_URL`) → **host OAuth proxy** → SuperGrok seats.2. `grok-rotate` / `auth.json` is the SuperGrok seat set; keep **auth-slots** in sync after reauth.3. Reauth egress: EFG SOCKS (`GROK_REAUTH_SOCKS`) → Bright Data tiers via`scripts/account-rotation/grok-reauth-egress.sh` (residential cascade).@@ -95,14 +95,14 @@ File: `$CLAUDE_PLUGIN_DATA_DIR/account-rotation/seat-state.json` (or `OPS_ACCOUN## Rules-1. Never print tokens, cookies, OTP codes, or vault dumps.-2. Never write real emails into committed files.-3. Prefer `bin/ops-accounts` over ad-hoc host paths.-4. Missing CRS is not a failure.-5. Dead RT → `reauth`, not “install CRS.”-6. Background long OAuth (Rule 4).+1. Never print tokens, cookies, OTP codes, or vault dumps.+2. Never write real emails into committed files.+3. Prefer `bin/ops-accounts` over ad-hoc host paths.+4. Missing cliproxy / CLIProxyAPI is not a failure.+5. Dead RT → `reauth`, not “install a relay.”+6. Background long OAuth (Rule 4).## Phase map-0 contract + this skill · 1 Claude parity · 2 CRS optional · 3 Grok complete ·-4 Codex+Factory · 5 Cursor · 6 companions · 7 host cutover · 8 gateway (no CRS)+0 contract + this skill · 1 Claude parity · 2 cliproxy optional · 3 Grok complete ·+4 Codex+Factory · 5 Cursor · 6 companions · 7 host cutover · 8 gateway (no relay required)diff --git a/claude-ops/skills/ops-desk/SKILL.md b/claude-ops/skills/ops-desk/SKILL.mdindex 4a6855e7..734f1637 100644--- a/claude-ops/skills/ops-desk/SKILL.md+++ b/claude-ops/skills/ops-desk/SKILL.md@@ -72,7 +72,7 @@ script encodes:tracker mutations from inside the workflow — they read, verify, and prepare only. Allexecution happens later in the main session under the outbound-approval gate.- **Batched fan-out (3 at a time).** A wide parallel burst can rate-limit (429) the- model-account pool — relay/CRS setups especially. Batches of 3 finish nearly as fast+ model-account pool — relay/CLIProxyAPI setups especially. Batches of 3 finish nearly as fastand never starve the rest of the fleet.- **Verification over trust.** Every load-bearing claim in a package must come from anactual read (issue body, thread, sqlite row) with dates/ids cited in `status_summary`.diff --git a/claude-ops/skills/ops-fleet/SKILL.md b/claude-ops/skills/ops-fleet/SKILL.mdindex 9a1753e0..86d564cd 100644--- a/claude-ops/skills/ops-fleet/SKILL.md+++ b/claude-ops/skills/ops-fleet/SKILL.md@@ -1,6 +1,6 @@---name: ops-fleet-description: Read-only fleet dashboard — every Claude session (local + remote) with token type (oauth/api/bedrock/crs), CRS relay state, account utilization, and per-session detail. No actions.+description: Read-only fleet dashboard — every Claude session (local + remote) with token type (oauth/api/bedrock/cliproxy), CLIProxyAPI relay state, account utilization, and per-session detail. No actions.argument-hint: '[--once] [--tui] [--all] [--no-color] [--no-ec2]'allowed-tools:- Bash@@ -13,22 +13,22 @@ maxTurns: 6Unified, read-only dashboard over the whole Claude session fleet — local `claude --bg`sessions plus any remote box reached over SSM — built from `claude agents --json`,-`claude daemon status`, the account-rotation daemon state, and the CRS (Claude Relay-Service) relay. It renders, it never mutates.+`claude daemon status`, the account-rotation daemon state, and the CLIProxyAPI relay+(legacy CRS-compatible deployments included). It renders, it never mutates.## Runtime ContextThe dashboard sources everything live; nothing needs to be parsed by this skill:- **Sessions**: `claude agents --json` (authoritative; nameless sessions fall back to their session id).-- **CRS relay**: health at `http://127.0.0.1:${CRS_PORT:-3005}/health`, allowlist at+- **CLIProxyAPI relay**: health at `http://127.0.0.1:${CRS_PORT:-3005}/health`, legacy allowlist at`~/.claude/scripts/account-rotation/crs-allowlist.json`, recent 429/529 from the relay window.- **Account pool**: the rotation daemon's keychain/util state (per-account util%, reset countdown).- **Remote (EC2/FRA) sessions**: a background-refreshed SSM cache (`~/.claude/state/fleet-ec2.json`,TTL ~45s); degrades to `refreshing…` / empty when unavailable.These paths are the standard account-rotator install locations-(`scripts/install-account-rotator-linux.sh`). If the rotator isn't installed the CRS /+(`scripts/install-account-rotator-linux.sh`). If the rotator isn't installed the CLIProxyAPI /account / remote rows simply degrade to `?` — the session table still renders.## Invocation@@ -44,14 +44,14 @@ Behavior:- With stdout captured (slash command / cron) it auto-falls back to a **one-shot snapshot**.- Flags: `--once`/`--snapshot` force a single render · `--tui`/`--watch [secs]` force the loop ·`--all` include completed sessions · `--no-color` plain · `--no-ec2` skip the remote SSM round-trip.-- Override the CRS port with `CRS_PORT=NNNN` (default 3005; 3000 is the legacy alias).+- Override the CLIProxyAPI-compatible relay port with `CRS_PORT=NNNN` (legacy env name; default 3005; 3000 is the legacy port alias).## OutputALWAYS render the dashboard to the user first — re-emit the snapshot inside a fenced codeblock (keep the box-drawing/bars/layout, strip ANSI). Below it add a one-line summary:-total sessions + token-type breakdown, anything blocked or flapping (high `att:`), CRS-health, and any account ≥90% util. The dashboard is the deliverable; the summary is the footnote.+total sessions + token-type breakdown, anything blocked or flapping (high `att:`), CLIProxyAPI+relay health, and any account ≥90% util. The dashboard is the deliverable; the summary is the footnote.End with a single line noting the live full-screen TUI is `${CLAUDE_PLUGIN_ROOT}/bin/ops-fleet`(or `--tui`) in a real terminal.diff --git a/claude-ops/skills/ops-mac/SKILL.md b/claude-ops/skills/ops-mac/SKILL.mdindex d323a493..3e7fb8cc 100644--- a/claude-ops/skills/ops-mac/SKILL.md+++ b/claude-ops/skills/ops-mac/SKILL.md@@ -110,7 +110,7 @@ For each actionable entry from `lanchr doctor` (missing binary, or repeated `exicom.example.foo — binary missing at [path]. This launchd job is dead.[Show the plist first] [Remove it] [Disable (keep file)] [Skip]```-Use `lanchr` to remediate where possible; otherwise `launchctl bootout`/`disable` after showing the plist. **Never touch a daemon the user relies on** (CRS/haproxy, gbrain push, ops-daemon, cloudflared tunnels, watchdogs) without spelling out exactly what it is — cross-check before recommending removal.+Use `lanchr` to remediate where possible; otherwise `launchctl bootout`/`disable` after showing the plist. **Never touch a daemon the user relies on** (CLIProxyAPI/legacy CRS/haproxy, gbrain push, ops-daemon, cloudflared tunnels, watchdogs) without spelling out exactly what it is — cross-check before recommending removal.### 3. Reclaimable disk```diff --git a/claude-ops/skills/ops-rotate/SKILL.md b/claude-ops/skills/ops-rotate/SKILL.mdindex 0f147675..4109c181 100644--- a/claude-ops/skills/ops-rotate/SKILL.md+++ b/claude-ops/skills/ops-rotate/SKILL.md@@ -1,6 +1,6 @@---name: ops-rotate-description: Alias of /ops:accounts for Claude Max seats (status, rotate-now, list, reauth, optional CRS). Prefer /ops:accounts for multi-provider. Full Claude procedure still in this file's historical detail via ops-accounts router.+description: Alias of /ops:accounts for Claude Max seats (status, rotate-now, list, reauth, optional cliproxy / CLIProxyAPI relay). Prefer /ops:accounts for multi-provider. Full Claude procedure still in this file's historical detail via ops-accounts router.argument-hint: '[status|rotate-now|list|add-account|reauth|crs|crs-tick]'allowed-tools:- Bashdiff --git a/claude-ops/templates/com.claude-ops.crs-401-refresher.plist b/claude-ops/templates/com.claude-ops.crs-401-refresher.plistindex d3953c8f..26b5f6d4 100644--- a/claude-ops/templates/com.claude-ops.crs-401-refresher.plist+++ b/claude-ops/templates/com.claude-ops.crs-401-refresher.plist@@ -1,6 +1,6 @@<?xml version="1.0" encoding="UTF-8"?><!--- CRS 401-refresher reconciler (claude-relay-service pool proactive OAuth+ CLIProxyAPI/legacy CRS-compatible 401-refresher reconciler (relay pool proactive OAuthtoken refresh).Schedule: every 300s. Equivalent systemd: OnUnitActiveSec=300s.diff --git a/claude-ops/templates/com.claude-ops.crs-429-cooldown.plist b/claude-ops/templates/com.claude-ops.crs-429-cooldown.plistindex a71ca73b..6a1942d6 100644--- a/claude-ops/templates/com.claude-ops.crs-429-cooldown.plist+++ b/claude-ops/templates/com.claude-ops.crs-429-cooldown.plist@@ -1,6 +1,6 @@<?xml version="1.0" encoding="UTF-8"?><!--- CRS 429-cooldown reconciler (claude-relay-service pool rate-limit hold).+ CLIProxyAPI/legacy CRS-compatible 429-cooldown reconciler (relay pool rate-limit hold).Schedule: every 60s. Equivalent systemd: OnUnitActiveSec=60s.The installer (scripts/install-crs-reconcilers-agent.sh) substitutesdiff --git a/claude-ops/templates/com.claude-ops.crs-priority.plist b/claude-ops/templates/com.claude-ops.crs-priority.plistindex aecc78e8..feedf22d 100644--- a/claude-ops/templates/com.claude-ops.crs-priority.plist+++ b/claude-ops/templates/com.claude-ops.crs-priority.plist@@ -1,6 +1,6 @@<?xml version="1.0" encoding="UTF-8"?><!--- CRS account-priority daemon (claude-relay-service pool auto-prioritization).+ CLIProxyAPI/legacy CRS-compatible account-priority daemon (relay pool auto-prioritization).Schedule: every 120s. Equivalent systemd: OnUnitActiveSec=120s.The installer (scripts/install-crs-priority-agent.sh) substitutes