cleanup

name	cleanup
description	Mechanical apply of /age's hash-anchored sidecar fixes via tilth_edit. The cleanup-wolf sub-agent is the only LLM in the path, fired per anchor mismatch.
license	MIT
compatibility	Requires Claude Code >= 2.1.30 / claude-agent-sdk >= 0.2.63 (tilth_edit must be exposed to plugin sub-agents).
metadata	{"owner":"cheese-flow","category":"cleanup"}
allowed-tools	["read","write","bash","subagent","mcp"]

Cleanup — Mechanical Fix Apply

Apply /age sidecar fixes mechanically via tilth_edit. No LLM in the happy path. The cleanup-wolf sub-agent fires only on hash mismatch.

Arguments

/cleanup <slug>

slug is the same slug produced by /age. Resolves to:

.cheese/age/<slug>.fixes.json

All harnesses share the project-root .cheese/ runtime directory.

Phase 1 — Load

Load .cheese/age/<slug>.fixes.json.

Validate schema: every entry must have all of id, dimension, file, anchor, content, rationale, category.

If any entry fails validation, abort with a structured error listing the invalid id values. Do not apply any fixes from a malformed file.

If the file does not exist, fail fast:

ERROR: .cheese/age/<slug>.fixes.json not found.
Run /age first to generate the fixes sidecar.

Phase 2 — Apply Each Fix

For each fix in fixes:

try:
  tilth_edit(
    path=fix.file,
    edits=[{
      start: fix.anchor.start,
      end:   fix.anchor.end,
      content: fix.content
    }]
  )
  record: applied fix.id

except HashMismatch:
  spawn cleanup-wolf sub-agent with:
    fix              — the original fix entry
    current_file     — cheez-read(fix.file, section containing anchor lines)
    rationale        — fix.rationale

  record wolf result: {id, status: "applied"|"skip"|"already_applied", ...}

Hash mismatch is the ONLY trigger for cleanup-wolf. Every other path is mechanical. Do not spawn the wolf for schema errors, missing files, or any other condition.

cleanup-wolf contract (see skills/cleanup/references/wolf-protocol.md):

The wolf receives the fix entry and the current file state at the affected region. It attempts to re-anchor via cheez-search narrative match and applies with tilth_edit. Returns one of:

{"id": "<fix.id>", "status": "applied", "new_anchor": {...}, "file": "..."}
{"id": "<fix.id>", "status": "skip", "reason": "<reason>"}
{"id": "<fix.id>", "status": "already_applied"}

Phase 3 — Emit Report

Write .cheese/age/<slug>.cleanup-report.md:

# Cleanup Report — <slug>

Applied:      <N> fixes
Wolf-rescued: <M> (hash mismatch, re-anchored and applied)
Skipped:      <K> (hash mismatch, wolf could not re-anchor)

## Skipped Details
<id> — <reason>

If all fixes were applied cleanly (zero wolf invocations), omit the Wolf-rescued and Skipped rows.

Print the report path to the caller.

Rules

• cleanup-wolf is the only LLM in this skill. The happy path has zero agent spawns.
• Do not modify <slug>.fixes.json or <slug>.suggestions.json.
• Do not apply suggestions. /cleanup is fixes-only; suggestions flow to /fromage cook.
• tilth_edit hash validation is the source of truth for anchor correctness. Do not second-guess it.
• Each fix is applied independently. A wolf failure on one fix does not block the remaining fixes.