| name | code-slimming-cleanup |
| description | Code slimming and cleanup guidelines. Use when reducing code size, removing dead code, simplifying abstractions, consolidating duplicate logic, or cleaning up after refactors while avoiding risky drive-by changes. |
| license | MIT |
Code Slimming and Cleanup
Guidelines for safely making code smaller, simpler, and cleaner without turning cleanup into an uncontrolled refactor.
Tradeoff: This skill favors safe, scoped cleanup over aggressive rewrites. If broader cleanup is needed, get explicit approval first.
Scan Procedure
Follow this order to avoid wasted work and ensure nothing is missed:
- Exclude — Identify build artifacts, generated files, and
node_modules/. Skip them from all subsequent steps.
- Hygiene — Check for missing
.gitignore, tracked build output, misplaced dependencies.
- File sizes — Count lines per source file. Flag files exceeding thresholds.
- Duplication — Grep for same-name function declarations across files. Compare bodies.
- Dead code — Search for unexported/unreferenced functions, unused imports, unreachable branches.
- Dependencies — Cross-reference
package.json entries against actual imports in source.
- Report — Output findings using the severity levels and format defined in this document.
For small projects (< 500 total source lines), skip steps 3-4 and only perform hygiene, dead code, and dependency checks. In the report, omit the Duplicates and Oversized files categories entirely rather than including empty sections.
Severity Levels
Every finding must be tagged with a severity:
| Level | Meaning | Examples |
|---|
| Critical | Wastes significant resources or blocks correct operation | Missing .gitignore with 300MB+ tracked artifacts; committed secrets |
| Warning | Maintenance burden or code smell that should be fixed soon | 100+ lines of exact duplication; file > 800 lines; unused dependency |
| Info | Worth noting but low urgency | File approaching 400 lines; 3-line utility duplicated twice; dependency that could move to devDependencies |
Order findings by severity in the report. Critical items first.
0. Exclude Build Artifacts and Generated Files
Never analyze or report generated files as cleanup candidates.
Before scanning, identify and skip:
- Build output directories (
dist/, build/, dist-electron/, .next/, .output/, out/)
- Bundled files (e.g., webpack/esbuild/rollup output like
assets/bundle.js)
- Compiled artifacts (
.exe, .dll, .asar, .pak, .bin)
- Auto-generated code (protobuf output, GraphQL codegen, OpenAPI clients)
node_modules/, vendor/, lock files
Instead, flag missing .gitignore rules if these artifacts are tracked in version control. The fix is a .gitignore entry, not deleting the files manually.
1. Define the Cleanup Scope
Name what kind of cleanup is being performed before editing.
Classify the task as one of:
- Local cleanup — remove unused imports, variables, branches, helpers, or files created by the current change. (Applies: sections 2, 3, 8, 9)
- Simplification — reduce unnecessary abstractions, indirection, configuration, or defensive code. (Applies: sections 2, 3, 5, 8, 9)
- Dead-code removal — delete code that is provably unreachable or unused. (Applies: sections 2, 3, 6, 8, 9)
- Duplication reduction — merge duplicate logic only when the shared behavior is truly identical. (Applies: sections 2, 4, 8, 9)
- Aggressive cleanup — broad deletion, reorganization, or public API changes. Requires explicit user approval. All sections apply; section 8's surgical constraint is relaxed but section 9 still applies.
If the task may affect behavior, public APIs, data formats, migrations, or external contracts, stop and confirm the scope.
2. Prefer Deletion Over Abstraction
Slimming means less code, not more structure.
When simplifying:
- Delete unnecessary code before introducing new helpers.
- Inline single-use abstractions when doing so improves readability.
- Remove unused options, flags, parameters, and configuration only when they are not part of a public contract.
- Do not add a framework, registry, strategy pattern, factory, or generic abstraction just to make cleanup look organized.
- Do not replace simple duplication with an abstraction unless the shared behavior is stable and obvious.
Ask: "Did this change reduce the amount of code a future reader must understand?"
3. Prove Code Is Dead Before Removing It
Do not delete code based on vibes.
Before removing existing code, verify at least one of:
- Type checker, compiler, linter, or IDE reports it as unused.
- Search shows no references across the relevant repository scope.
- Tests cover the path and still pass after removal.
- The code is behind a removed feature path or obsolete configuration explicitly identified in the task.
- The user explicitly asked to remove that code.
If unsure, mention the suspected dead code instead of deleting it.
4. Detect and Reduce Duplication
Identical logic repeated across files is a maintenance liability, not just a style issue.
Detection strategy:
- Grep for function declarations with the same name across all source files.
- For each match, compare function bodies — classify as "exact" (identical) or "near" (identical AST structure with only identifier renames or argument-vs-closure differences; anything involving different control flow, different literals, or different called functions is NOT a near-match).
- If bodies differ in meaningful ways (different edge-case handling, different return types), they are NOT duplicates — leave them alone.
When scanning for duplication:
- Flag functions with identical or near-identical bodies appearing in 2+ files.
- Threshold: 10+ lines of duplicated logic warrants extraction to a shared module.
- Only merge when the behavior is truly identical — similar-looking code with subtle differences should stay separate.
- Prefer a flat utility module over deep abstraction hierarchies.
Runtime boundary awareness: If duplication exists because of browser/Node isolation (e.g., browser code cannot import a Node module that uses fs or zlib), do NOT recommend "just import the Node version." Instead:
- Extract the pure-logic subset (no platform APIs) into a shared module that both environments can import.
- If the shared module would be trivial (< 5 lines), note the duplication as Info severity and move on.
Report duplicates with: function name, file locations, line counts, and whether they are exact or near-matches.
5. Enforce File Size Limits
Large files are harder to review, test, and maintain.
Flag files that exceed these thresholds:
| Metric | Warning | Action required |
|---|
| Single file lines | > 400 | Flag for review |
| Single file lines | > 800 | Recommend splitting |
| Single function lines | > 80 | Flag for extraction |
| Single function lines | > 150 | Recommend splitting |
When recommending a split:
- Identify natural seams (e.g., separate concerns, independent state, distinct API surface).
- Prefer splitting by responsibility, not by arbitrary line count.
- Do not split if the file is a single cohesive unit (e.g., a binary format encoder/decoder).
6. Audit Dependencies
Unused or redundant dependencies add install time, attack surface, and bundle size.
Check for:
dependencies or devDependencies not imported anywhere in source.- Dependencies duplicating built-in functionality (e.g.,
lodash.get when optional chaining suffices).
- Dependencies only used in one place that could be trivially inlined.
dependencies that should be devDependencies (build tools, test frameworks, bundlers).
Classifying dependencies vs devDependencies:
- If a package is
imported or required at runtime → dependencies.
- If a package is only used by build scripts, test runners, or bundlers →
devDependencies.
- If a package is not imported at runtime but its static files (e.g.,
dist/) are referenced by a bundler config (pkg.assets, build.files, copy-webpack-plugin) → it must stay in dependencies for the bundler to resolve it. Do not move it to devDependencies.
Tools: npx depcheck, npx knip, or manual grep for package names in src/.
For non-Node projects, substitute equivalent tooling: Python → pip-check/vulture; Go → go mod tidy/staticcheck; Rust → cargo-udeps. The dependency rules apply to the equivalent manifest file (e.g., requirements.txt, go.mod, Cargo.toml).
7. Check Version Control Hygiene
Tracked files that should be ignored waste repo size and cause noise.
Flag if any of these are committed:
- Build output (
dist/, build/, dist-electron/, .next/, .output/, out/)
- Bundled assets that have a build script (e.g.,
assets/bundle.js when npm run build generates it)
- OS/editor files (
.DS_Store, Thumbs.db, .idea/, .vscode/ with user-specific settings)
- Secrets or credentials (
.env, *.pem, credentials.json)
Recommend adding a .gitignore if one is missing. If artifacts are already tracked, note that git rm --cached is needed to stop tracking without deleting local files.
8. Keep Cleanup Surgical
Do not turn cleanup into a drive-by refactor.
Allowed:
- Remove code made unused by your current change.
- Simplify code directly involved in the requested task.
- Rename only when the old name becomes misleading because of the cleanup.
- Update tests directly affected by removed or simplified behavior.
Avoid unless explicitly requested:
- Reformatting unrelated files.
- Moving files for aesthetics.
- Changing public APIs.
- Rewriting working code in a preferred style.
- Mixing unrelated cleanup items in one diff.
- Deleting comments or docs you do not fully understand.
The test: every deleted or changed line should have a clear reason tied to the cleanup goal.
9. Preserve Behavior Unless Asked Otherwise
A slimming PR should usually be behavior-preserving.
Before and after cleanup:
- Identify expected behavior that must stay the same.
- Run existing tests when available.
- Add or update tests only when needed to lock down behavior that cleanup could accidentally change.
- If tests are missing, use type checks, builds, search, or targeted manual checks as fallback verification.
If the cleanup intentionally changes behavior, state that clearly and treat it as a feature or bug-fix change, not pure cleanup.
10. Report Format
When finishing, output findings in this structure:
### [Severity] Category heading
| Location | Lines | Issue | Suggestion |
|----------|-------|-------|------------|
| file:line | N | description | action |
Group by severity (Critical → Warning → Info), then by category within each severity.
Required categories in the report (omit any category with no findings rather than including an empty section):
- Hygiene issues: missing
.gitignore, tracked build artifacts, misplaced dependencies.
- Deletable code: files, functions, branches, imports proven dead or redundant.
- Duplicates: functions to consolidate, with file locations, line counts, match type.
- Oversized files: files/functions exceeding thresholds, with split suggestions.
- Dependency issues: unused, redundant, or misclassified packages.
- Kept intentionally: suspicious code that looked removable but was not proven safe to delete.
- Verification needed: what tests, builds, or manual checks are required before acting.
End with a one-line summary quantifying the total impact, e.g.:
Summary: 3 critical, 5 warning, 2 info. Estimated reduction: ~320 source lines + 380 MB tracked artifacts.
Good cleanup leaves reviewers confident that less code now does the same job.
