IronLint Init

SkillDev tools

Bootstraps a project's .ironlint.yml with IronLint's generic baseline, then helps the user deliberately add checks for their stack and existing linters. Use when user says "init ironlint", "set up ironlint", "bootstrap ironlint config", "create ironlint checks", "ironlint init", or asks to create or generate an ironlint configuration.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the IronLint Init skill

What this skill tells your AI

The instructions your AI receives, as published by ironlint/ironlint in adapters/claude-code/skills/ironlint-init/SKILL.md and read by ahel’s review.

Start with IronLint's generic baseline, then help the user choose checks for their stack and installed linters. A check is two fields — files (a glob or list) and run (a shell command that exits nonzero to block); there are no engines or severities.

This skill is user-driven. Do not silently install tools or add checks. Every step below is a proposal the user accepts or declines.

Step 1: Run ironlint init

ironlint init writes a small, stack-agnostic starter .ironlint.yml and trusts it for you. It does not inspect manifest files or wrap linters:

ironlint init

This creates .ironlint.yml with generic starter checks (and, outside this Claude session, can also wire hooks into other agents — not needed here, the PreToolUse hook is already running). Review the generated checks.

Step 2: Wrap the project's linters as checks

For each linter the user has installed (ruff, biome, eslint, tsc, phpstan, clippy, …) that runs per file, propose a check that feeds the proposed content on stdin; any nonzero exit blocks:

checks:
  ruff-check:
    files: ["**/*.py"]
    run: "ruff check --quiet --stdin-filename \"$IRONLINT_FILE\" -"

Test each candidate against a sample file before adding it (see /ironlint-config for the fixture loop). Only add checks that pass on clean input and block on dirty input. Skip repo-wide tools that aren't per-file (e.g. cargo clippy) — they don't map to a per-file check; suggest running them as a pre-push step instead.

Step 3: Trust the config

ironlint init already trusted the config it scaffolded. If you hand-edit .ironlint.yml (Step 2), re-bless it:

ironlint trust

This records a sha256 of the config (and any files it extends:/.ironlint/scripts/) in the out-of-repo trust store at ~/.config/ironlint/trust.json — it does not write into .ironlint.yml. Any later edit invalidates the fingerprint, and ironlint check refuses to run until you re-trust.

Step 4: Verify

Edit any in-scope file. The PreToolUse hook runs ironlint and either passes (clean) or blocks (with the check's message). See the ironlint skill for how to read a block verdict.

Notes

  • If .ironlint.yml already exists, do not overwrite it — ironlint init leaves an existing config untouched. Propose edits via /ironlint-config instead.
  • There is no migration from older formats; ironlint rejects a pre-0.3 config (schema_version:/rules:) outright. Write checks fresh.
  • Telemetry lands at .ironlint/log.jsonl, with a per-check breakdown in each record. The /ironlint-review skill consumes it.

Signals

GitHub stars
20
Forks
2
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
ironlint-init
Source
github.com/ironlint/ironlint