Code Bloat Stinger

Code Bloat Stinger

Designed and built by Legion Code Inc.

Where this skill's files are

Every path below (guides/, references/, scripts/) is relative to the folder that contains this SKILL.md. In Claude Code that folder is ${CLAUDE_SKILL_DIR}. If that text was not replaced with a real path, find the folder yourself (Glob for **/code-bloat-stinger/SKILL.md) and use it as SKILL_DIR in every command.

Requirements

Node 20.19 or newer, Python 3.9 or newer, and Perl for cloc or a downloaded scc binary (Go or Cargo only for those repos). The commands in references/tool-commands.md are bash (env -i, bash arrays, [ ] tests): on Windows, run them in Git Bash or WSL, not PowerShell or cmd. The throwaway copy (audit-my-code-stinger/scripts/make_scratch_copy.py) and this skill's scripts/*.py are plain Python and run anywhere.

Purpose

Answer "how bloated is this codebase, and why?" with numbers a founder can trust. The skill measures three kinds of bloat separately, because they have different causes and fixes: copy-pasted code (jscpd at two clone sizes, with both the standard and the both-copies figure), unused code (knip for JS and TS, vulture, deadcode, staticcheck or cargo tools elsewhere, always checked by hand before it is called dead), and shape (share of code in files over 1,000 lines, the largest functions, factory closures, and a split plan). It writes the bloat dimension of an Audit My Code run (01-bloat/) under the shared contract, and never changes the audited repository: every tool runs in a throwaway copy that is removed afterwards. The method reproduces a hand-run reference audit (WHIPT, 2026-09-30) on any repository.

When to use

• "Is this codebase bloated?", "is it bloated because it was vibe coded?", "how much is copy-pasted?", "how much dead code is there?"
• The bloat dimension of /audit-my-code at the vibe, basic, full, or mario tier
• Measuring duplication, unused files, exports and dependencies, file and function size, or planning how to split giant files
• Re-running the same measures after a cleanup to show what changed

When not to use

• Security weaknesses: security-stinger and the framework security Drones (02-security/)
• Whether the code matches its plan, tests, or quality bar: quality-stinger (03-quality/)
• Repo setup, CI, docs, and hygiene: the foundation Drone (00-foundation/)
• Actually deleting code or refactoring: this skill only reports; remediation belongs to the mario tier and the owning engineering Drone
• Dependency vulnerability triage: dependency-audit-stinger

Procedure

1. Read guides/00-scoping-and-throwaway-copy.md. Read intake.json, make the copy with git archive, install pinned tools outside the copy, write the scope and file list, capture git status of the audited repo, pick the language route.
2. Run duplication per guides/01-duplication-with-jscpd.md: 50 and 100 tokens, size and line caps lifted, scan coverage checked against the file list, then scripts/jscpd_recount.py for both copies, per area, big versus small files, and top pairs.
3. Run unused-code detection per guides/02-unused-code-with-knip.md: workspace-aware knip from outside the repo, a failing plugin disabled through an external config, a production-mode second view, then scripts/knip_summarize.py. For non-JS code use the tools in references/non-js-dead-code.md.
4. Count lines per guides/03-size-and-line-counts.md with cloc (--skip-uniqueness) or scc, then scripts/size_profile.py.
5. Measure structure per guides/04-structure-and-split-planning.md with lizard and scripts/function_profile.py; read the largest files and write the split plan, including the context-object step for closures.
6. Spot-check per guides/05-spot-check-and-false-positives.md: at least 12 flagged items classified Dead, False alarm, or Partial with a reason, then the truly-dead estimate as a share of core code lines.
7. Band, set the verdict, write the verdict paragraph, and write metrics.json, findings.jsonl, and notes.md per guides/06-bands-verdict-and-contract-output.md. List raw artifacts under ## Evidence in notes.md (the orchestrator builds evidence-index.md). Run scripts/validate_outputs.py and fix every error.
8. Confirm the audited repo is unchanged, remove .scratch/, and report to the orchestrator: verdict, headline measures, caps lifted, and anything not run.

References map

Related drones and stingers

• code-bloat-wasp-drone - The paired Drone; the orchestrator dispatches it for the bloat dimension.
• audit-my-code-stinger - The run contract and orchestration for /audit-my-code.
• security-stinger - Security dimension; overlaps only where dead code hides a vulnerable path.
• quality-stinger - Quality dimension, run after security in the full and mario tiers.
• dependency-audit-stinger - Vulnerable or outdated dependencies, as opposed to unused ones.

Critical Directive

• You must read all files and context contained within your skill.
• In the event your core knowledge does not provide sufficient guidance you must make every attempt to search the internet, related knowledge base documentation files, and other available resources to supplement your knowledge prior to proceeding with your task.
• Additional related skills can be found here:
  • audit-my-code-stinger - Run contract, tiers, workspace layout, and the brief renderer for Audit My Code.
  • webapp-capture-stinger - Code-level UI inconsistency audit, when bloat findings point at duplicated components.

Designed and built by Legion Code Inc. Licensed under AGPL-3.0-or-later.