toolchain-healthlisted

# Toolchain Health — verify the assumptions before they bite Every field failure this team has hit traces to an *assumed* tool that wasn't there: Playwright skipped (the two-gradient UI shipped "verified"), graphify off the PATH (agents fell back to grep), a canvas server that died. This skill checks the assumptions up front — evidence, not hope. ## Fits in the pipeline - **Stage 0/5 (recommended):** run at `/kickoff` and before the Stage-5 go-for-code gate. - **On any tool failure:** a failed render, graph query, or hook — diagnose here first. - Owned by `toolchain-doctor` (Reliability/Ops). Checked by `claim-verifier` (see below). ## How to run it Run the bundled doctor from the PROJECT directory (it probes the project's own setup): ``` node "${CLAUDE_PLUGIN_ROOT}/skills/toolchain-health/tools/doctor.mjs" ``` Exit 0 = healthy (missing OPTIONAL tools don't fail) · exit 1 = something configured/required is BROKEN. It checks: **Node ≥18 · git · Claude CLI · Playwright resolvable from the project (the v0.8.0 render mandate) · graphify** (a `.graphify-path` marker that doesn't run = BROKEN → re-run `/graph`; absent graphify = optional) **· mermaid-cli (optional) · VIBEGOD_GUARDRAILS mode**. ## Discipline 1. **Paste the actual table** — the verdict is the script's output, never a summary from memory. 2. **Fix forward:** each ✗/– line carries a FIX command. Offer to run the fixes (installs need the user's yes); re-run the doctor after, and only then report healthy. 3. **Mak