Agent-almanac clean-codebase

install
source · Clone the upstream repo
git clone https://github.com/pjt222/agent-almanac
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/pjt222/agent-almanac "$T" && mkdir -p ~/.claude/skills && cp -r "$T/i18n/caveman-ultra/skills/clean-codebase" ~/.claude/skills/pjt222-agent-almanac-clean-codebase-5b9b71 && rm -rf "$T"
manifest: i18n/caveman-ultra/skills/clean-codebase/SKILL.md
source content

clean-codebase

Use When

Codebase has hygiene debt:

  • Lint warns piled up during rapid dev
  • Unused imports + vars clutter files
  • Dead code paths never removed
  • Formatting inconsistent across files
  • Static analysis reports fixable issues

Do NOT use for architectural refactor, bug fixes, or business logic changes. This = hygiene + automated cleanup only.

In

ParamTypeRequiredDescription
codebase_path
stringYesAbsolute path to codebase root
language
stringYesPrimary language (js, python, r, rust, etc.)
cleanup_mode
enumNo
safe
(default) or
aggressive
run_tests
booleanNoRun test suite after cleanup (default: true)
backup
booleanNoCreate backup before deletion (default: true)

Do

Step 1: Pre-Cleanup Assessment

Measure current state → quantify gains later.

# Count lint warnings by severity
lint_tool --format json > lint_before.json

# Count lines of code
cloc . --json > cloc_before.json

# List unused symbols (language-dependent)
# JavaScript/TypeScript: ts-prune or depcheck
# Python: vulture
# R: lintr unused function checks

Baseline metrics saved to

lint_before.json
+
cloc_before.json

If err: Lint tool not found → skip automated fixes, manual review

Step 2: Fix Automated Lint Warnings

Apply safe auto fixes (spacing, quotes, semis, trailing ws).

JavaScript/TypeScript:

eslint --fix .
prettier --write .

Python:

black .
isort .
ruff check --fix .

R:

Rscript -e "styler::style_dir('.')"

Rust:

cargo fmt
cargo clippy --fix --allow-dirty

All safe lint warns resolved; files formatted consistent

If err: Auto fixes break tests → revert, escalate

Step 3: Identify Dead Code Paths

Static analysis → unreferenced fns, unused vars, orphaned files.

JavaScript/TypeScript:

ts-prune | tee dead_code.txt
depcheck | tee unused_deps.txt

Python:

vulture . | tee dead_code.txt

R:

Rscript -e "lintr::lint_dir('.', linters = lintr::unused_function_linter())"

General approach:

  1. Grep fn defs
  2. Grep fn calls
  3. Report fns defined but never called

dead_code.txt
lists unused fns, vars, files

If err: Static analysis tool unavail → manual review recent commit history for orphans

Step 4: Remove Unused Imports

Clean import blocks → drop refs to pkgs never used.

JavaScript:

eslint --fix --rule 'no-unused-vars: error'

Python:

autoflake --remove-all-unused-imports --in-place --recursive .

R:

# Manual review: grep for library() calls, check if package used
grep -r "library(" . | cut -d: -f2 | sort | uniq

All unused imports removed

If err: Removing imports breaks build → used indirectly → restore + doc

Step 5: Remove Dead Code (Mode-Dependent)

Safe Mode (default):

  • Remove code explicit marked deprecated
  • Remove commented-out blocks (>10 lines + >6 months old)
  • Remove TODO comments for completed issues

Aggressive Mode (opt-in):

  • Remove all unused fns from Step 3
  • Remove private methods w/ zero refs
  • Remove feature flags for deprecated features

Each candidate deletion:

  1. Valid. zero refs in codebase
  2. Check git history → skip if modified last 30 days
  3. Remove + add entry to
    CLEANUP_LOG.md

Dead code removed;

CLEANUP_LOG.md
documents all deletions

If err: Uncertain code truly dead → move to

archive/
dir vs. delete

Step 6: Normalize Formatting

Consistent formatting all files (even if linters miss).

  1. Normalize line endings (LF vs CRLF)
  2. Single newline at EOF
  3. Remove trailing ws
  4. Normalize indentation (spaces vs tabs, width)
# Example: Fix line endings and trailing whitespace
find . -type f -name "*.js" -exec sed -i 's/\r$//' {} +
find . -type f -name "*.js" -exec sed -i 's/[[:space:]]*$//' {} +

All files follow consistent formatting conventions

If err: sed breaks binary files → skip + doc

Step 7: Run Tests

Valid. cleanup didn't break functionality.

# Language-specific test command
npm test              # JavaScript
pytest                # Python
R CMD check           # R
cargo test            # Rust

All tests pass (or same fails as pre-cleanup)

If err: Revert incrementally → identify breaking change → escalate

Step 8: Generate Cleanup Report

Doc all changes for review.

# Codebase Cleanup Report

**Date**: YYYY-MM-DD
**Mode**: safe | aggressive
**Language**: <language>

## Metrics

| Metric | Before | After | Change |
|--------|--------|-------|--------|
| Lint warnings | X | Y | -Z |
| Lines of code | A | B | -C |
| Unused imports | D | 0 | -D |
| Dead functions | E | F | -G |

## Changes Applied

1. Fixed X lint warnings (automated)
2. Removed Y unused imports
3. Deleted Z lines of dead code (see CLEANUP_LOG.md)
4. Normalized formatting across W files

## Escalations

- [Issue description requiring human review]
- [Uncertain deletion moved to archive/]

## Validation

- [x] All tests pass
- [x] Backup created: backup_YYYYMMDD/
- [x] CLEANUP_LOG.md updated

Report saved to

CLEANUP_REPORT.md
in project root

If err: (N/A — generate report regardless)

Check

Post-cleanup:

  • All tests pass (or same fails as before)
  • No new lint warns introduced
  • Backup created pre-delete
  • CLEANUP_LOG.md
    documents all removed code
  • Cleanup report generated w/ metrics
  • Git diff reviewed for unexpected changes
  • CI pipeline passes

Traps

  1. Remove Code Still Used via Reflection: Static analysis misses dynamic calls (e.g.,

    eval()
    , metaprogramming). Always check git history.

  2. Break Implicit Deps: Removing imports used by deps. Run tests after every import removal.

  3. Delete Feature Flags for Active Features: Unused in current branch, but maybe active in other envs. Check deployment configs.

  4. Over-Aggressive Formatting: Tools like

    black
    /
    prettier
    reformat → unnecessary diffs. Configure tools → project style.

  5. Ignore Test Coverage: Can't safely clean codebases w/o tests. Low coverage → escalate for test additions first.

  6. No Backup: Always create

    backup_YYYYMMDD/
    dir pre-delete, even w/ git.

  7. Wrong R binary on hybrid systems: WSL / Docker,

    Rscript
    maybe resolves to cross-platform wrapper vs. native R. Check w/
    which Rscript && Rscript --version
    . Prefer native R binary (e.g.,
    /usr/local/bin/Rscript
    Linux/WSL) for reliability. See Setting Up Your Environment for R path config.