π The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
TypeScript agentic-aiagentic-frameworkagentic-workflowagentsai-agentsai-assistantai-skillsautonomous-agentsclaudecodexdsh-pluginharnessmcp-servermulti-agentmulti-agent-systemsnpmskillsswarmswarm-intelligencetypescript
chore(security): document Socket.dev alert baseline + remove broken pages.yml workflow (#2340)
rUv16a55f7a537c
diff --git a/.github/workflows/pages.yml b/.github/workflows/pages.ymldeleted file mode 100644index d51f647e51..0000000000--- a/.github/workflows/pages.yml+++ /dev/null@@ -1,25 +0,0 @@-name: Deploy Pages-on:- push:- branches: [main]- paths:- - 'docs/**'--permissions:- pages: write- id-token: write--jobs:- deploy:- environment:- name: github-pages- url: ${{ steps.deployment.outputs.page_url }}- runs-on: ubuntu-latest- steps:- - uses: actions/checkout@v4- - uses: actions/configure-pages@v4- - uses: actions/jekyll-build-pages@v1- with:- source: ./docs- - uses: actions/deploy-pages@v4- id: deploymentdiff --git a/docs/security/socket-baseline.md b/docs/security/socket-baseline.mdnew file mode 100644index 0000000000..8b4a0edeea--- /dev/null+++ b/docs/security/socket-baseline.md@@ -0,0 +1,87 @@+# Socket.dev Alert Baseline β `claude-flow`++> Last revised: 2026-06-09 against `[email protected]`+> Source: <https://socket.dev/npm/package/claude-flow/alerts/3.10.40>+> Tracking issue: ruvnet/ruflo#2339++## TL;DR++Most of Socket's alerts on `claude-flow` describe **legitimate behaviour of a CLI agent platform** (network access, filesystem access, env-var reads, shell access, native modules, etc.) and cannot and should not be "fixed" β they are intrinsic to what the package does. This doc separates the inherent-and-expected categories from the small set that's actually actionable, so the next reviewer doesn't go down the same rabbit hole.++## What's protected today++| Layer | Mechanism | Status |+|---|---|---|+| Ruflo dev tree | `overrides` block in root `package.json` (lines 77-107) β pins `protobufjs >=8.2.0`, `uuid >=14.0.0`, plus 25+ other transitive forces | β `npm audit` returns 0 vulnerabilities on the working tree |+| Ruflo dev tree | `supply-chain-audit` CI job in `.github/workflows/v3-ci.yml` (line 629) β runs `npm audit --audit-level=high`, lockfile integrity, top-level allowlist, typosquat reject, publisher trust snapshot | β Gates every PR + push to main |+| Dependency review | `dependency-review` CI job (line 675) β GitHub's `dependency-review-action` checks new vulnerable deps in PRs | β οΈ Gated on Pages-style Dependency Graph being enabled at repo level; currently `continue-on-error: true` |++## What's NOT fully fixable from inside `claude-flow`++`npm overrides` in our root `package.json` only apply when **we** are the root of the dep tree. When a consumer runs `npm install [email protected]`, *their* root `package.json` is the one whose overrides apply β ours are ignored. This is the same trap documented in CLAUDE.md for the `ruflo` wrapper (#2112).++A clean install of `[email protected]` therefore still pulls in:++| Severity | Package | CVE summary | Root cause |+|---|---|---|---|+| Critical | `protobufjs` | Arbitrary code execution via bytes-field defaults in generated `toObject` code | Transitive of `onnx-proto` β `onnxruntime-web` β `@xenova/transformers` |+| Moderate | `uuid` | Missing buffer bounds check in v3/v5/v6 when `buf` is provided | Transitive β multiple paths |+| High (cascade) | `onnx-proto`, `onnxruntime-web`, `@xenova/transformers`, `agentdb`, `agentic-flow`, `@claude-flow/memory`, `@claude-flow/neural`, `@claude-flow/plugin-gastown-bridge`, `claude-flow` itself | All downstream of the protobufjs + uuid root causes | β |++**Why this can't be fixed in this PR**: the upstream `@xenova/[email protected]` (latest) still pins `[email protected]` which carries the bad `onnx-proto` β `protobufjs` chain. No clean upstream version exists today. Resolving consumer-side requires either:++1. A patched `@xenova/transformers` release (waiting on upstream), or migrating to `@huggingface/transformers` (the rebranded successor β needs evaluation), OR+2. Republishing the affected `@claude-flow/*` sub-packages with explicit transitive pins in *their* `dependencies` blocks (multi-PR coordinated effort), OR+3. Dropping the optional ML stack (`agentic-flow`, `@xenova/transformers`) β would lose ONNX-backed features++Tracked in #2339 as separate follow-up work.++## Inherent flags β what they mean, why we keep them++Socket fires these on every release. They describe normal CLI/agent behaviour and are not signals of a bug.++| Alert | Count | Why it's normal here |+|---|---|---|+| Network access | 68 pkgs | HTTP clients (Anthropic SDK, MCP transports, `fetch`-based tools) |+| Filesystem access | 82 pkgs | Universal β anything reading/writing files (memory backend, config, logs) |+| Environment variable access | 84 pkgs | `process.env.ANTHROPIC_API_KEY`, `CLAUDE_FLOW_*`, etc. β configuration |+| Shell access | 29 pkgs | `child_process` β used by every native-module installer, hook runners, `npx`-style spawning |+| Install scripts | 7 pkgs | Native modules: `better-sqlite3`, `hnswlib-node`, `onnxruntime-node`, `@ruvector/*` |+| Native code | 5 pkgs | Compiled binaries from the install-scripts row |+| URL strings | 108 pkgs | Every HTTP client embeds endpoint URLs as string literals |+| Debug access | 14 pkgs | The `debug` package and reflection-based helpers |+| Dynamic require | 16 pkgs | Plugin system (`mcp`, hooks, agents) + conditional optional-dep loading |+| Minified code | 10 pkgs | Some libraries ship pre-bundled (`undici`, `axios` internals, etc.) |+| Unpopular package | 29 pkgs | Long-tail transitives β popularity is not a security signal |+| Unmaintained | 64 pkgs | Long-tail transitives stale >5 years β checked against CVE/deprecation separately |+| New author | 20 pkgs | Maintainer changes β informational, not a vuln |++## False positives we've explicitly triaged++| Alert | What Socket said | Reality |+|---|---|---|+| AI-detected possible typosquat β "did you mean `z-schema`?" | Suggested `zod` is a typosquat of `z-schema` | `zod` is a 10M-weekly-download canonical TypeScript schema library, not a typosquat. Socket's AI guess is wrong. |+| AI-detected potential security risk | 15 instances in 7 packages | Heuristic-only; none corroborated by an actual CVE or code review |+| AI-detected potential code anomaly | 52 instances in 46 packages | Same β heuristic-only |+| Obfuscated code | 17 instances in 11 packages | All match the "pre-bundled / minified" pattern (`undici`, `axios` internals); none match the active-obfuscation-malware pattern |++## Operational policy++1. **Root tree must stay green.** The `supply-chain-audit` job in `v3-ci.yml` enforces `npm audit --audit-level=high` on every PR. Adding a new direct dep that introduces a HIGH/CRITICAL fails CI.+2. **Overrides are a first-line tool.** If a new transitive CVE lands, add an `overrides` entry pinning to a patched version. The current set (root `package.json` lines 77-107) is the precedent.+3. **Document gaps before merging.** When a CVE can't be cleanly overridden (e.g., the protobufjs cascade documented above), update this file with the rationale rather than silently ignoring it.+4. **Re-baseline this doc when claude-flow versions bump.** The alert counts here are tied to `3.10.40` β when the version moves, re-run the audit (`scripts/probe-nested-spawn-depth.mjs` style) and update the tables.++## When to revisit++- **Quarterly**: re-check upstream for `@xenova/transformers` / `onnxruntime-web` patched versions+- **On Socket alert change**: if Socket flags something new at *critical* severity for a direct dep, that's a real signal and warrants immediate triage+- **On new direct dep addition**: the `supply-chain-audit` job will catch this at PR time; no manual checklist needed++## Related++- ADR-097 β federation budget circuit-breaker (cost-side supply chain)+- ADR-145 β plugin supply-chain integrity (install-time)+- ADR-144 β authorization propagation (action-time, ADR-131 is content-time)+- Issue #2046 β original supply-chain hardening (the source of `supply-chain-audit` job)+- Issue #2339 β this baseline + Socket alert response