forge-auditlisted

# forge-audit Detect and fix drift between the six context files and the real codebase. Over a long build the code moves on; stale context files quietly make the agent guess wrong. This skill keeps them honest. ## Argument Text after the command narrows the audit (e.g. `/forge-audit architecture` or `/forge-audit just the budgets`) — run only the matching sections below. No argument → full audit. ## First: read the state Run the deterministic detector to know exactly what exists before auditing (read-only): ```bash bash "${CLAUDE_PLUGIN_ROOT}/skills/forge-init/scripts/detect.sh" ``` If the `verdict` is `SETUP` (no context files), there's nothing to audit — tell the user to run `forge-init` first. Otherwise proceed. ## How to audit Read the six files in `context/`, then compare each against evidence in the repo. Do NOT trust the docs — verify against the code. For the evidence gathering, delegate to the `forge-scout` agent (haiku-pinned, read-only): give it the concrete claims extracted from each context file (mission "drift evidence") and let it report confirmed/drifted/gone with file:line evidence — the sweep stays out of this session's context. Judging the findings and recommending doc edits stays here. If the agent is unavailable, verify in-session. ### architecture.md - Re-read `package.json` / lockfile / equivalent. Compare the declared stack table to the dependencies actually installed. Flag tech listed that's gone, and tech in use that's undocumented. -