carrylint

One of three zero-dependency linters for AI-agent repos. To run all three in a single pass, with one report and one exit code, use tenken — npx @hyuga/tenken.

Your skill works on your machine. Does it work on your teammate's — or in a different agent? carrylint is a zero-dependency, model-agnostic linter that fails your PR when a SKILL.md, AGENTS.md, or slash-command has your machine or your model baked in — absolute paths, undeclared external CLIs, unresolved placeholders. It runs no LLM and needs no API key: pure static analysis.

あなたのスキル、自分のマシンでは動く。でも "次の人の環境・別のエージェント" で動きますか？ carrylint は、SKILL.md / AGENTS.md / スラッシュコマンドに作った本人の環境・モデル前提が焼き込まれていないかを CI で落とす、依存ゼロ・モデル非依存のリンタ。絶対パス・未宣言の外部CLI・未解決プレースホルダを検出します。実行時に LLM も API キーも使いません（純静的解析）。

Why / なぜ

Agent Skills became an open standard in Dec 2025 — one SKILL.md runs across 20+ agents (Claude Code, Codex, Gemini CLI, Cursor, Copilot …). The format is portable now. But the content still isn't: a skill that shells out to codex you never installed, writes to C:\Users\you\Downloads, or hardcodes gpt-image-2 silently breaks the moment someone else installs it.

Other skill linters check your skill is spec-valid. carrylint checks it actually runs when someone else installs it.

オープン標準化で "形式" は可搬になった。carrylint は "中身が実際に動くか" を見る側です。

Tried on real code / 実データに当てた

3.8% of a random sample of 2,465 skills published on ClawHub contain an absolute path that resolves only on the author's machine — /Users/root009/projects/demos/g1/game2, D:\Personal\OpenClaw\figures\. Drawn from 69,265 enumerated, August 2026, seed 20260804.

The same rule fires on the first-party skills bundled in openclaw/openclaw (385k stars), where openclaw-live-updater instructs the agent to treat /Users/steipete/openclaw — the maintainer's own home directory — as a deployment mirror.

Precision is treated as the product. v0.1.1 was retuned after a 230-repository audit cut its false positives by ~85%; v0.2.2 stopped reading /home/YOUR_USER/… as an author path while deliberately still catching /Users/CS/…, which is a real person's initials. Both directions are pinned in test/realworld.test.mjs.

What it checks / 何を見るか

| kind | severity | 説明 | |---|---|---| | abs-path | error | 作者環境前提の絶対パス（C:\Users\… / /Users/<実名>/ / /home/<実名>/）。$HOME / ~ / %USERPROFILE% と汎用名（/home/user 等）は可搬なので対象外 | | placeholder | error | 配布物に残った未完成マーカー（<FILL_ME> / REPLACE_ME / CHANGEME / <INSERT …>）。YOUR_API_KEY / /path/to/ は「置き換えてね」の正当な文書慣習なので対象外 | | undeclared-cli | warn | 外部/プロバイダCLI（codex ollama …）を宣言なしで叩く。ホストの設定/確認（claude mcp add codex --version 等）は除外 | | provider-env | warn | プロバイダ固有 API キー env の生参照（OPENAI_API_KEY …） | | todo | warn | 配布物に残った TODO: / FIXME: | | model-id | opt-in | モデルID直書き（claude-* / gpt-* / gemini-*）※--model-ids で有効化 |

Low-noise by design: only unambiguous, author-only breakers are error (fail the PR); the rest are warn. Validated against a real-world audit of 230 public skills (v0.1.1): every remaining error was a genuine hardcoded personal path.

誤検知＝狼少年化が唯一の死因。実データ230リポの監査で ERROR の誤検知をゼロに追い込みました（v0.1.1）。

Use as a GitHub Action / CIで使う（定着の本体）

# .github/workflows/carrylint.yml
name: carrylint
on: [push, pull_request]
jobs:
  carrylint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: hyuga611/carrylint@v0
        with:
          paths: .          # optional; default = repo
          # strict: 'true'      # warnings also fail
          # model-ids: 'true'   # enable the opt-in model-id rule

Findings appear as inline PR annotations and the job fails (exit 1) on any error, so a non-portable skill can't be merged.

Use as a CLI / ローカルで使う

npx @hyuga/carrylint                     # scan the repo
npx @hyuga/carrylint path/to/skills      # scan a dir/file
npx @hyuga/carrylint --allow codex,gemini   # I depend on these on purpose
npx @hyuga/carrylint --strict            # warnings fail too
npx @hyuga/carrylint --model-ids         # enable the model-id rule
npx @hyuga/carrylint --format json       # machine-readable (VS Code / tooling)

Suppress a line / 行単位の無効化

Save to `C:\tools\out.png` <!-- carry-ignore -->

<!-- carry-ignore-next -->
This model is pinned on purpose: `claude-opus-4-8`

Config / 設定

The provider-CLI / model-ID / env dictionary lives in rules.json (data-driven — edit it, no code change). CLI flags also read env: CARRYLINT_STRICT=1, CARRYLINT_MODEL_IDS=1, CARRYLINT_FORMAT=json.

Model-agnostic / モデル非依存

carrylint treats codex / gemini / claude / ollama / aider … equally as provider lock-in — it never favors one vendor. The check itself is "are you locked to one machine or one model?", so it works the same whether you author with Claude, Codex, or Gemini. No LLM at runtime.

How it differs / 既存との違い

• reflint — does the reference exist? (referential integrity)
• skills-lint — do two skills collide? is the frontmatter valid?
• carrylint — does the reference resolve on someone else's machine / another agent? (runtime portability)

Part of a small family of zero-dependency, language-agnostic, CI-resident linters for AI-native artifacts.

License

Related tools

Zero-dependency CI linters for repos where AI agents do the work. Each one fails the PR on something that breaks quietly.

| | Catches | | --- | --- | | tenken — start here | Runs reflint + skills-lint + carrylint over one tree: one report, one exit code, one Action | | reflint | AGENTS.md / llms.txt / CLAUDE.md pointing at commands, scripts, or paths that no longer exist | | skills-lint | SKILL.md broken references + name/trigger collisions between skills | | carrylint ← you are here | Skills with the author's machine or model baked in — absolute paths, undeclared CLIs, unresolved placeholders | | genchi | Agents reporting "done" without re-fetching real-world state | | tracklint | Forms and CTAs that quietly stopped being wired for conversion tracking | | tokenlint | Hardcoded colors that bypass your design tokens | | reflint for VS Code | The same reflint checks, inline in the editor as you save | | orogami | Not a linter — natural Japanese/CJK line breaking for OGP images (BudouX + font subsetting) |

MIT © hyuga611