Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write-time code quality enforcement using Plankton — auto-formatting, linting, and Claude-powered fixes on every file edit via hooks.
.claude/skills/loulanyue-plankton-code-quality/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 125% | 0% |
Integration reference for Plankton (credit: @alxfazio), a write-time code quality enforcement system for Claude Code. Plankton runs formatters and linters on every file edit via PostToolUse hooks, then spawns Claude subprocesses to fix violations the agent didn't catch.
Every time Claude Code edits or writes a file, Plankton's multi_linter.sh PostToolUse hook runs:
Phase 1: Auto-Format (Silent)
├─ Runs formatters (ruff format, biome, shfmt, taplo, markdownlint)
├─ Fixes 40-50% of issues silently
└─ No output to main agent
Phase 2: Collect Violations (JSON)
├─ Runs linters and collects unfixable violations
├─ Returns structured JSON: {line, column, code, message, linter}
└─ Still no output to main agent
Phase 3: Delegate + Verify
├─ Spawns claude -p subprocess with violations JSON
├─ Routes to model tier based on violation complexity:
│ ├─ Haiku: formatting, imports, style (E/W/F codes) — 120s timeout
│ ├─ Sonnet: complexity, refactoring (C901, PLR codes) — 300s timeout
│ └─ Opus: type system, deep reasoning (unresolved-attribute) — 600s timeout
├─ Re-runs Phase 1+2 to verify fixes
└─ Exit 0 if clean, Exit 2 if violations remain (reported to main agent)| Scenario | Agent sees | Hook exit | |----------|-----------|-----------| | No violations | Nothing | 0 | | All fixed by subprocess | Nothing | 0 | | Violations remain after subprocess | [hook] N violation(s) remain | 2 | | Advisory (duplicates, old tooling) | [hook:advisory] ... | 0 |
The main agent only sees issues the subprocess couldn't fix. Most quality problems are resolved transparently.
LLMs will modify .ruff.toml or biome.json to disable rules rather than fix code. Plankton blocks this with three layers:
protect_linter_configs.sh blocks edits to all linter configs before they happenstop_config_guardian.sh detects config changes via git diff at session end.ruff.toml, biome.json, .shellcheckrc, .yamllint, .hadolint.yaml, and moreA PreToolUse hook on Bash blocks legacy package managers:
pip, pip3, poetry, pipenv → Blocked (use uv)npm, yarn, pnpm → Blocked (use bun)npm audit, npm view, npm publish> Note: Plankton requires manual installation from its repository. Review the code before installing.
bash# Install core dependencies brew install jaq ruff uv # Install Python linters uv sync --all-extras # Start Claude Code — hooks activate automatically claude
No install command, no plugin config. The hooks in .claude/settings.json are picked up automatically when you run Claude Code in the Plankton directory.
To use Plankton hooks in your own project:
.claude/hooks/ directory to your project.claude/settings.json hook configuration.ruff.toml, biome.json, etc.)| Language | Required | Optional | |----------|----------|----------| | Python | ruff, uv | ty (types), vulture (dead code), bandit (security) | | TypeScript/JS | biome | oxlint, semgrep, knip (dead exports) | | Shell | shellcheck, shfmt | — | | YAML | yamllint | — | | Markdown | markdownlint-cli2 | — | | Dockerfile | hadolint (>= 2.12.0) | — | | TOML | taplo | — | | JSON | jaq | — |
| Concern | ECC | Plankton | |---------|-----|----------| | Code quality enforcement | PostToolUse hooks (Prettier, tsc) | PostToolUse hooks (20+ linters + subprocess fixes) | | Security scanning | AgentShield, security-reviewer agent | Bandit (Python), Semgrep (TypeScript) | | Config protection | — | PreToolUse blocks + Stop hook detection | | Package manager | Detection + setup | Enforcement (blocks legacy PMs) | | CI integration | — | Pre-commit hooks for git | | Model routing | Manual (/model opus) | Automatic (violation complexity → tier) |
If running both ECC and Plankton hooks:
Plankton's .claude/hooks/config.json controls all behavior:
json{ "languages": { "python": true, "shell": true, "yaml": true, "json": true, "toml": true, "dockerfile": true, "markdown": true, "typescript": { "enabled": true, "js_runtime": "auto", "biome_nursery": "warn", "semgrep": true } }, "phases": { "auto_format": true, "subprocess_delegation": true }, "subprocess": { "tiers": { "haiku": { "timeout": 120, "max_turns": 10 }, "sonnet": { "timeout": 300, "max_turns": 10 }, "opus": { "timeout": 600, "max_turns": 15 } }, "volume_threshold": 5 } }
Key settings:
volume_threshold — violations > this count auto-escalate to a higher model tiersubprocess_delegation: false — skip Phase 3 entirely (just report violations)| Variable | Purpose | |----------|---------| | HOOK_SKIP_SUBPROCESS=1 | Skip Phase 3, report violations directly | | HOOK_SUBPROCESS_TIMEOUT=N | Override tier timeout | | HOOK_DEBUG_MODEL=1 | Log model selection decisions | | HOOK_SKIP_PM=1 | Bypass package manager enforcement |
Set strict quality behavior:
bashexport ECC_HOOK_PROFILE=strict export ECC_QUALITY_GATE_FIX=true export ECC_QUALITY_GATE_STRICT=true
During quality enforcement, flag changes to config files in same iteration:
biome.json, .eslintrc*, prettier.config*, tsconfig.json, pyproject.tomlIf config is changed to suppress violations, require explicit review before merge.
Use the same commands in CI as local hooks:
Track:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 17,949 | 16,296 | -9% | 1 | 1 | 0% | 3,018 | 5,402 | +79% | 0 | 0 | — |
case-02 | fail→pass | 16,140 | 11,844 | -27% | 1 | 1 | 0% | 2,800 | 4,446 | +59% | 0 | 0 | — |
case-03 | fail→pass | 19,962 | 19,121 | -4% | 1 | 1 | 0% | 3,380 | 5,525 | +63% | 0 | 0 | — |
case-04 | pass→pass | 16,030 | 11,639 | -27% | 1 | 1 | 0% | 2,949 | 4,508 | +53% | 0 | 0 | — |
case-05 | fail→pass | 14,304 | 11,039 | -23% | 1 | 1 | 0% | 2,603 | 4,151 | +59% | 0 | 0 | — |
case-06 | pass→fail | 11,635 | 8,360 | -28% | 1 | 1 | 0% | 2,000 | 3,713 | +86% | 0 | 0 | — |
case-07 | fail→pass | 7,602 | 2,386 | -69% | 1 | 1 | 0% | 1,145 | 2,575 | +125% | 0 | 0 | — |
case-08 | pass→pass | 7,426 | 2,136 | -71% | 1 | 1 | 0% | 1,112 | 2,534 | +128% | 0 | 0 | — |
case-09 | fail→pass | 7,388 | 2,484 | -66% | 1 | 1 | 0% | 1,049 | 2,596 | +147% | 0 | 0 | — |
case-10 | fail→pass | 5,691 | 1,577 | -72% | 1 | 1 | 0% | 781 | 2,416 | +209% | 0 | 0 | — |
case-11 | fail→pass | 8,379 | 3,539 | -58% | 1 | 1 | 0% | 1,259 | 2,724 | +116% | 0 | 0 | — |
case-12 | fail→pass | 12,336 | 1,694 | -86% | 1 | 1 | 0% | 1,875 | 2,456 | +31% | 0 | 0 | — |
case-13 | fail→pass | 13,441 | 2,147 | -84% | 1 | 1 | 0% | 2,163 | 2,512 | +16% | 0 | 0 | — |
case-14 | fail→pass | 18,820 | 1,689 | -91% | 1 | 1 | 0% | 3,081 | 2,398 | -22% | 0 | 0 | — |
case-15 | fail→pass | 7,274 | 1,675 | -77% | 1 | 1 | 0% | 1,146 | 2,404 | +110% | 0 | 0 | — |
case-16 | pass→pass | 5,216 | 2,634 | -50% | 1 | 1 | 0% | 807 | 2,531 | +214% | 0 | 0 | — |
case-17 | fail→pass | 9,474 | 1,585 | -83% | 1 | 1 | 0% | 1,509 | 2,420 | +60% | 0 | 0 | — |
case-18 | fail→pass | 4,592 | 1,912 | -58% | 1 | 1 | 0% | 680 | 2,465 | +263% | 0 | 0 | — |
case-19 | fail→pass | 24,605 | 3,032 | -88% | 1 | 1 | 0% | 2,274 | 2,391 | +5% | 0 | 0 | — |
case-20 | fail→pass | 15,507 | 2,225 | -86% | 1 | 1 | 0% | 2,400 | 2,566 | +7% | 0 | 0 | — |
case-21 | fail→pass | 19,872 | 1,834 | -91% | 1 | 1 | 0% | 1,023 | 2,465 | +141% | 0 | 0 | — |
case-22 | pass→pass | 3,646 | 1,485 | -59% | 1 | 1 | 0% | 565 | 2,371 | +320% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +73 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.