| name | ant-colony |
| description | Run the Ant code-cleanup colony (scout/fix/review/apply) and report findings, verified fixes, and skips. Use when the user asks to scan for code issues, run Ant, clean up unused imports / dead code / N+1 queries, or wants a CI-style code-health check. |
| argument-hint | [scout|fix|review|apply] [path] [extra ant flags] |
| allowed-tools | Bash(node *) Bash(ant *) |
Ant colony skill
A THIN front door over the ant binary. This skill runs ant <verb> --json
as a subprocess and parses its newline-delimited event stream into a readable
summary. It contains NO detection, fixing, trust, or orchestration logic — all
of that lives in the Go engine (TECHSPEC §3). The skill only execs the binary
and renders what the --json stream already says happened.
Run
node ${CLAUDE_SKILL_DIR}/src/run.ts $ARGUMENTS
Instructions
The block above has already run ant <verb> --json and printed a structured
summary. Using only that output:
- Lead with the headline: findings count, verified, skipped, applied, and the
highest severity seen.
- List each finding (species, file:line, severity, message).
- For every SKIP, surface the failing verifier and its reason — a skip is a
trust signal, never a hidden error. Do not downplay skips.
- For verified/applied fixes, name the fixer (provenance) and the verifiers
that passed.
- If the run reported an error (operational failure such as a missing detector
binary), report it plainly and stop.
Do not invent findings, re-run detection, or make fix/trust decisions — the
engine owns those. Render only what the stream contains.
