ThumbGate Doctor

Audit whether ThumbGate is actually wired into this agent — PreToolUse / SessionStart hooks installed, MCP server reachable, lesson store present, statusline, overall agent-readiness — then tell the user exactly what to fix.

This skill wraps existing ThumbGate capability and adds no new logic — it runs the existing doctor + integrity checks.

Workflow

1. Run the wiring/health audit:

   npx thumbgate doctor

(Add --json for a machine-readable report.) It exits non-zero when the project is not ready.

1. Verify the runtime path with the check_operational_integrity MCP tool — this confirms the

server-side enforcement path is live, not just the local config.

1. Summarize: ✅ what's wired (hooks, MCP, store, statusline) and ❌ what's missing, each with

its exact fix command (usually npx thumbgate init).

1. Lead with the fix. If everything is green, say so plainly with the readiness status; if not,

lead with the single highest-impact fix.

The full readiness checklist (each check, what a failure means, and its fix) is in references/wiring-checklist.md.

Example

Input: "why aren't my gates firing? is ThumbGate even wired up?"

Action: run npx thumbgate doctor, then check_operational_integrity, then:

❌ PreToolUse hook not installed — run npx thumbgate init. ✅ MCP server reachable. ✅ Lesson store present. Fix the hook and your promoted rules will start blocking.

Troubleshooting

• `npx thumbgate doctor` exits non-zero: that's the point — read which checks failed and apply

the printed fix (usually npx thumbgate init).

• Doctor green but gates still don't fire: the rule may not be active — check the thumbgate-rules

skill — or no gated action has been attempted yet — check the thumbgate-blocked skill.

• `check_operational_integrity` unreachable while doctor is green: local config is fine but the

MCP server isn't running — restart it (npx thumbgate serve).

Quality checklist (self-verify before delivering)

• [ ] I ran the real npx thumbgate doctor and reported its actual readiness status, not a guess.
• [ ] I confirmed the runtime path with check_operational_integrity, not just local config.
• [ ] I listed each ❌ with its exact fix command (e.g. npx thumbgate init).
• [ ] If green, I said so with the readiness status; if not, I led with the single highest-impact fix.
• [ ] I added no new logic — only ran the existing doctor + integrity checks.