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.
claude-ops-installer
Install, update, and verify the claude-ops plugin across Claude Code, Codex, Gemini CLI, OpenClaw, Hermes, and OpenCode from one command.
git clone https://github.com/Lifecycle-Innovations-Limited/claude-ops.git
cd claude-ops/installer
npm install
node bin/claude-ops-installer.mjs install
Check first: is every CLI on the same skills? (read-only)
On a machine that already has OPS installed, this answers in one command whether Claude Code, Codex, Grok, Cursor and Hermes load byte-identical OPS skills. It writes, fetches and locks nothing.
node bin/claude-ops-installer.mjs check --source claude-installed
--source claude-installed uses the release Claude Code has installed as the
reference. To compare against a tagged release instead, fetch it into the
installer cache first (the only step that downloads), then check:
node bin/claude-ops-installer.mjs fetch --ref v3.10.28
node bin/claude-ops-installer.mjs check --ref v3.10.28
check --source <dir> needs no npm dependencies. The other subcommands need
npm install; without it they stop with an environment error (exit 5), not a
drift or auth error. Statuses, every exit code and the fix for each case are in
the plugin's docs/skill-parity.md.
What it does
Reads the canonical source (Lifecycle-Innovations-Limited/claude-ops at a pinned ref) and mirrors skills + scripts + binstubs into each detected agent's expected layout. One central config governs all agents. Re-run any time to refresh after an upstream release.
Supported agents (day-1)
| Agent | Strategy | Default path |
|---|---|---|
| Claude Code | Marketplace install (or symlink fallback) | ~/.claude/plugins/cache/ops-marketplace/ops/current/ |
| Codex | Flat ln -s |
~/.codex/skills |
| Gemini CLI | Flat ln -s |
~/.gemini/skills |
| OpenClaw | Flat ln -s |
~/.openclaw/skills |
| Hermes | Hybrid skills + native plugin | ~/.hermes/skills and ~/.hermes/plugins/ops |
| OpenCode | Flat ln -s |
~/.config/opencode/skills |
Binstubs from upstream bin/ are symlinked into ~/bin/ (or $CLAUDE_OPS_BIN_DIR).
Subcommands
claude-ops-installer check [--source <dir>|claude-installed] [--ref <ref>] [--host a,b] [--require a,b] [--paths] [--json]
claude-ops-installer fetch [--ref <ref>]
claude-ops-installer install [--ref <ref>] [--agents a,b,c] [--dry-run] [--force]
claude-ops-installer update [--ref <ref>] [--agents a,b,c]
claude-ops-installer verify [--agents a,b,c]
claude-ops-installer doctor [--agents a,b,c]
claude-ops-installer uninstall [--agents a,b,c]
claude-ops-installer agents
claude-ops-installer --help
| Flag | Effect |
|---|---|
--ref <ref> |
Git ref: tag, branch, or sha. Default from config. |
--agents a,b,c |
Limit which agents get touched. Default: all enabled. |
--dry-run |
Print the planned actions, change nothing. |
--force |
Kept for compatibility. A real file/dir at a target is never deleted: it is reported as OWNERSHIP_CONFLICT. The installer only replaces symlinks it owns. |
--source <dir> |
check only: reference plugin root, or claude-installed. |
--host a,b |
check only: diagnostic selection. Required hosts left out report NOT_CHECKED. |
--require a,b |
check only: required hosts. Default: required in ~/.config/claude-ops/parity.json, else none. |
--offline |
check only: skip the read-only git ls-remote that reports a stale cache. |
verify, doctor and check never fetch or create the cache. install,
update and fetch are the only subcommands that download, and they key the
cache by the resolved commit SHA, so a cache for a moving ref is reported as
STALE_SOURCE instead of passing silently. A failed fetch keeps the existing
cache and names the error class. One writer per target root: a second run gets
LOCKED (exit 6).
| --json | Emit machine-readable JSON instead of human text. |
Central config
Default path: ~/.config/claude-ops-installer/config.yaml (XDG). Falls back to ~/.claude-ops-installer.yaml.
version: 1
source:
type: git
url: https://github.com/Lifecycle-Innovations-Limited/claude-ops.git
ref: v3.12.0
agents:
claude: { enabled: true }
codex: { enabled: true, path: ~/.codex/skills }
gemini: { enabled: true, path: ~/.gemini/skills }
openclaw: { enabled: true, path: ~/.openclaw/skills }
hermes:
{
enabled: true,
flat: ~/.hermes/skills,
nested: ~/.hermes/skills/ops,
plugin: ~/.hermes/plugins/ops,
}
opencode: { enabled: false, path: ~/.config/opencode/skills }
bin:
path: ~/bin
strategy: symlink # or copy
Override per call: --agents codex,hermes ignores the config's enabled flag for this invocation.
Public-repo rule (the operator 2026-07-21)
This package ships in a public repo (Lifecycle-Innovations-Limited/claude-ops). It MUST NOT contain:
- Real names, emails, phone numbers, or usernames (use
<[email protected]>) - Real store URLs, project names, or org names (use
<yourstore.myshopify.com>,<my-project>) - API keys, tokens, secrets, session strings, or chat IDs
- Real GitHub org or repo slugs in examples
- Hardcoded paths like
/Users/<name>/...(use~or$HOME)
All user-specific data lives in the central config file, which is gitignored by convention. CI runs tests/test-no-secrets.sh to verify before merge.
Exit codes
One table for the installer, sync-companion-clis.sh and the parity check
(claude-ops/lib/parity/status.json; a test fails if this table drifts from it):
| Code | Class | Meaning |
|---|---|---|
| 0 | clean |
Everything matched or applied and verified |
| 1 | drift |
Bytes differ, or a target points at a developer checkout |
| 2 | gap |
A required target or the reference is missing, stale or unverifiable |
| 3 | partial |
Some targets applied, others failed or were refused; the failed ones kept their previous state |
| 4 | usage |
Bad arguments, or no agents enabled |
| 5 | environment |
Missing dependency, invalid config, unexpected runtime error |
| 6 | locked |
Another OPS run holds the target lock |
Earlier versions documented 2 = fetch failed and 3 = config invalid; neither was ever returned (both failures exited 1). 4 for "no agents enabled" is unchanged.
License
MIT — see ../LICENSE.