Clean code

Code Cleanup — Orchestrator

Productivity — Multi-agent orchestration. Audits and refactors existing code in-place for readability, maintainability, and dead-code removal — without changing behavior. Produces a dated cleanup record.

Core Question: "Is this change purely structural with zero behavioral impact?"

Why this skill exists, methodology, principles, when NOT to refactor, history: references/playbook.md [PLAYBOOK].

When To Use

• Codebase has accumulated dead code, AI slop, unused dependencies, or production-waste assets.
• After major feature additions, before release milestones, when test runtime grows.
• When onboarding new team members and structural cruft slows them down.
• Standalone — no upstream gate required.

When NOT To Use

• Cleanup is mixed with a feature change (separate commits — always).
• No test coverage AND behavior-preservation matters (write tests first; see references/anti-patterns.md [ANTI-PATTERN] "When NOT to refactor").
• Pre-existing test/build failures unrelated to cleanup (BLOCKED until baseline is green).
• Code that won't change again — if nobody will read or modify it, the investment doesn't pay off.
• Duplication is the same operational mechanics repeated across 2+ callers (identical SDK setup, retry loop, I/O plumbing) → use /extract-service. clean-code removes dead code and slop; it does not design service boundaries.

Critical Gates (The 5 Golden Rules)

Before delivering, the critic-agent verifies ALL golden rules pass:

1. Preserve behavior — Every change must produce the same observable behavior. If you can't verify this, don't make the change.
2. Small incremental steps — One change at a time. Commit between steps. Never combine a refactor with a feature change.
3. Check existing conventions first — Read the codebase's existing coding guidelines, linting config, naming patterns, and file structure. Match them.
4. Test after each change — Run the test suite after every modification. If tests break, revert and try a smaller step.
5. Rollback awareness — Commit before starting. Note the hash. If a change chain gets too complex, revert and try a different approach.

Additional gate: Session limits — target ~30 changes per cleanup session. After 15 changes, generate an interim summary. If each fix spawns 2+ new issues, stop and reassess.

If any golden rule fails: the critic identifies the specific change that violated it and recommends reverting. Never silently bypass — the rules are the safety contract. Full failure-handling flow: references/anti-patterns.md [ANTI-PATTERN] "When the critic FAILs."

Safety supersedes --fast: all 5 rules fire under --fast, single-agent fallback, and dry-run modes (per mode-resolver safety-gates-supersede contract). Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.

Before Starting

Apply the before-starting-check [PLAYBOOK]:

Step	Action
0	Mode resolution — budget: deep. Mode-resolver (references/_shared/mode-resolver.md [PROCEDURE]) auto-downgrades to fast for ≤5-file scopes (→ Single-Agent Fallback in dispatch-mechanics); --fast flag forces single-agent. Safety gates supersede --fast.
1	Read implementation-roadmap/canonical-paths.md if present — verify output path matches canonical inventory.
2	Read .forsvn/index/manifest.json for prior cleanup runs against the same scope; surface staleness if recent cleanup already covered this path.
3	Read docs/forsvn/experience/technical.md for prior conventions notes.

Pre-Dispatch

Run the Pre-Dispatch protocol (references/_shared/pre-dispatch-protocol.md).

Needed dimensions: codebase path, cleanup intent (dead code / unused deps / asset / refactor / mixed), test suite available, conventions to preserve.

Read order: (1) codebase scan — package manifest, test config, lint config, framework hints (CLAUDE.md, .editorconfig); (2) docs/forsvn/experience/technical.md for prior conventions notes.

Warm Start (obvious intent), Cold Start (vague invocation), and write-back rules: references/pre-dispatch-prompts.md [PROCEDURE].

Routing + Dispatch

Multi-agent orchestration (8 agents across 2 layers: 4 parallel scanners → safe-removal → refactoring → validation → critic), triage rules, dispatch protocol, routing-rules table, and single-agent fallback: references/procedures/dispatch-mechanics.md [PROCEDURE]. Load at Layer 1 dispatch entry.

For an annotated full-codebase walkthrough (Express API, all 4 scanners + Layer 2 + critic decisions): references/examples/cleanup-walkthrough.md [EXAMPLE].

Artifact Contract

• Path: docs/forsvn/artifacts/meta-clean-code-<YYYY-MM-DD>-cleanup-<slug>.md (flat v2 grammar; re-run same slug same day → append .v[N]). Stack is meta (cleanup records are meta-stack snapshots, like diagnose); skill is in product/ because consumers are product-side.
• Lifecycle: snapshot (dated, immutable record of one cleanup run).
• Frontmatter fields: skill, version, date, status (DONE / DONE_WITH_CONCERNS / BLOCKED / NEEDS_CONTEXT), stack (=meta), review_surface (=none — snapshot defaults to decision_state: not_required), lifecycle, produced_by, provenance. v2 schema: references/_shared/artifact-contract-template.md.
• Required sections: Scope, Changes Made (≥1 subsection populated), Validation, Critic Verdict. Manual Verification Needed + Rollback when applicable.
• Consumed by: clean-artifacts (scans filenames for staleness), review-work (when reviewing cleanup-touched code), operator (history audit).
• Full template: references/report-template.md [PROCEDURE].

Chain Position

Previous: none | Next: none (standalone).

Re-run triggers: after major feature additions, before release milestones, when test runtime grows significantly, when onboarding new team members.

Anti-Patterns

Critic-load reference: references/anti-patterns.md [ANTI-PATTERN]. Re-read before applying any change that smells off — large batch, behavioral side-effect, untested deletion, convention override, generated-code touch. "When NOT to refactor" exit conditions also live there.

Next Step

Run /review-work for a fresh-eyes quality review on cleanup-touched code.

Durable Rules (protected)

<!-- SLOW_UPDATE_START --> <!-- No pinned rules yet. Populate via the slow-update workflow (see references/slow-update-fence.md). Each pinned rule must (a) be procedural not instance-specific, (b) be earned from a regression or critic-flagged failure, (c) cite the artifact / decision record that justified pinning. --> <!-- SLOW_UPDATE_END -->

Completion Status

Every run ends with explicit status:

• DONE — all approved removals applied, behavior preserved (tests + lint + build PASS), critic PASS.
• DONE_WITH_CONCERNS — cleanup applied but some validation skipped (no test suite, pre-existing build break, manual verification required); report flags what wasn't checked.
• BLOCKED — pre-existing test/build failures unrelated to cleanup; pause so the baseline can be fixed before proceeding (otherwise rollback signal is unreliable).
• NEEDS_CONTEXT — codebase conventions unclear (no framework detected, mixed language stack, ambiguous test runner); ask user before scanning.

References

• references/playbook.md [PLAYBOOK] — why, methodology, principles, when NOT to refactor, history
• references/procedures/dispatch-mechanics.md [PROCEDURE] — agent roster, execution layers, dispatch protocol, routing rules, single-agent fallback
• references/_shared/{pre-dispatch-protocol, before-starting-check, mode-resolver, artifact-contract-template}.md — canonical shared specs
• references/pre-dispatch-prompts.md [PROCEDURE] — Warm + Cold prompts verbatim
• references/ai-slop-patterns.md — code-scanner pattern catalog
• references/production-waste-patterns.md — asset-scanner pattern catalog
• references/anti-patterns.md [ANTI-PATTERN] — failure modes + When NOT to refactor + When the critic FAILs
• references/report-template.md [PROCEDURE] — artifact frontmatter + section template + filename conventions
• references/examples/cleanup-walkthrough.md [EXAMPLE] — Express API cleanup end-to-end
• scripts/analyze_codebase.py — static analysis tool used by structural-scanner, dependency-scanner, asset-scanner (junk files, empty dirs, large dirs, unused code, unused/broken/duplicate assets, unoptimized media)