forge-doctor

Runs Forge's self-test + secret-leak scan across source, tests, and dashboard. Use before shipping or when asked to test everything, check it executes, or find leaks.

SKILL.md

GitHub

forge-doctor — self-test + leak scan

One command that proves the Forge install in this project actually works and leaks nothing.

When to use

  • The owner asks to "test everything", "check it all executes", or "find leaks".
  • Before a sync/handoff, or after a batch of tooling changes.
  • As the Security Boss's evidence step (the leak scan is a real security check).

How to run

node .claude/forge-bin/forge-doctor.cjs                 # print a green/red summary, exit 0/1
node .claude/forge-bin/forge-doctor.cjs --run <run_id>  # also write <run>/doctor.json + log doctor_run
node .claude/forge-bin/forge-doctor.cjs --json          # full machine-readable report

What it checks (all real — no fabrication)

  1. node --check on every .cjs/.js under forge-bin/ + forge-dashboard/.
  2. tests — runs every forge-bin/*.test.cjs and tallies passed/failed.
  3. honesty gate — feeds log-event.cjs a known type (expect accept) and an unknown type (expect hard-reject / exit 2), then deletes the throwaway run. Proves strict events still reject junk.
  4. dashboard SPA — the render files are all present.
  5. leak scan — scans git-tracked files with the hardened SECRET_PATTERNS; reports only {file, pattern}, never the matched secret. .env.example + binaries are skipped.

Writes <run>/doctor.json (read by the dashboard Doctor panel + the Project Registry's test_status). Exit code 0 = all green, 1 = a check failed — safe to gate a hook/CI on it.

Honesty

Real command output only. A failing check is reported as failing; the leak scan never prints a secret, only the file + which pattern matched. test_status in the registry mirrors this doctor.json.