Default Steps

Airlock ships with a set of built-in steps in the airlock-hq/airlock/defaults/ namespace. These handle the most common workflow tasks.

`rebase`

Reference: airlock-hq/airlock/defaults/rebase@main

Rebases onto the upstream branch to handle drift before running the rest of the pipeline.

- name: rebase
  uses: airlock-hq/airlock/defaults/rebase@main

`lint`

Reference: airlock-hq/airlock/defaults/lint@main

Runs linters and formatters on the codebase, auto-fixing issues where possible.

Detects your project's linter configuration (ESLint, Prettier, Biome, Ruff, etc.)
Applies auto-fixes directly to the worktree (must run pre-freeze)
Produces patch artifacts for any changes made

- name: lint
  uses: airlock-hq/airlock/defaults/lint@main

`critique`

Reference: airlock-hq/airlock/defaults/critique@main

Critiques the code change for bugs, risks, and simplification opportunities using AI.

Produces comment artifacts anchored to specific files and lines
Writes critique_result.json with max_severity (none, info, warning, error) and comment_count
Always exits 0 — results are in the artifact, not the exit code

- name: critique
  uses: airlock-hq/airlock/defaults/critique@main

`test`

Reference: airlock-hq/airlock/defaults/test@main

Runs your project's test suite using an AI agent.

Detects your test runner (Jest, Vitest, pytest, go test, etc.)
Produces content artifacts with test results summary
Writes test_result.json with verdict (pass, fail, skip) and summary
Always exits 0 — the verdict is in the artifact, not the exit code. Use a downstream gate step to pause on failure.

- name: test
  uses: airlock-hq/airlock/defaults/test@main

`gate`

Reference: airlock-hq/airlock/defaults/gate@main

Evaluates implementation risk using AI and conditionally pauses for human approval.

Reads test_result.json and critique_result.json from upstream steps
Uses an AI agent to assess risk based on test results, critique findings, change scope, and implementation ambiguity
Assigns a risk level: low, medium, or high
Compares the risk level against AIRLOCK_RISK_THRESHOLD to decide whether human approval is required
Writes risk_assessment.json with the assessment details
Produces a content artifact summarizing the risk assessment

- name: review
  uses: airlock-hq/airlock/defaults/gate@main
  env:
    # Change to never, low, medium, or high to control when human approval is required.
    AIRLOCK_RISK_THRESHOLD: medium

Risk Levels

Level	Score	Meaning
`low`	1	Change is well-bounded, mostly cosmetic or straightforward, with little ambiguity
`medium`	2	Change has meaningful behavioral impact, notable uncertainty, or concerning signals
`high`	3	Change is fundamental, risky, ambiguous, or has strong negative signals from tests/critique

`AIRLOCK_RISK_THRESHOLD`

Controls the threshold at which the gate pauses for human approval. If the assessed risk score is greater than or equal to the threshold score, the pipeline pauses with airlock exec await.

Value	Behavior
`never`	Never pause for human approval
`low`	Pause for all non-trivial changes (risk >= low)
`medium`	Pause for risky changes (risk >= medium) — default
`high`	Pause only for high-risk changes

`describe`

Reference: airlock-hq/airlock/defaults/describe@main

Generates a PR title and description from the diff using AI.

Analyzes the diff between AIRLOCK_BASE_SHA and AIRLOCK_HEAD_SHA
Produces content artifacts (title and description)
Produces comment artifacts (inline review comments)
Typically runs post-freeze

- name: describe
  uses: airlock-hq/airlock/defaults/describe@main

`document`

Reference: airlock-hq/airlock/defaults/document@main

Updates documentation to reflect the changes in the diff.

Analyzes the diff and existing docs to identify what needs updating
Produces patch artifacts for documentation changes
Typically runs post-freeze, after describe

- name: document
  uses: airlock-hq/airlock/defaults/document@main

`push`

Reference: airlock-hq/airlock/defaults/push@main

Pushes the validated code to the upstream remote.

Reads AIRLOCK_UPSTREAM_URL and pushes AIRLOCK_BRANCH
Only runs after all validations and approvals
Does nothing if the Push Request was canceled

- name: push
  uses: airlock-hq/airlock/defaults/push@main

`create-pr`

Reference: airlock-hq/airlock/defaults/create-pr@main

Creates a pull request (or merge request) on the upstream host.

Uses the title and description from content artifacts produced by earlier steps
Attaches comment artifacts as PR review comments
Supports GitHub, GitLab, and Bitbucket

- name: create-pr
  uses: airlock-hq/airlock/defaults/create-pr@main

Default Pipeline Structure

The default workflow runs steps in parallel where possible:

Wave 1: [rebase]
Wave 2: [critique] [test]          ← parallel
Wave 3: [gate]                     ← conditionally pauses for approval
Wave 4: [describe] [document]      ← parallel
Wave 5: [lint → push → create-pr]  ← sequential deploy

Critique and test run in parallel after rebase. The gate step reads their result artifacts, performs an AI-driven risk assessment, and calls airlock exec await if the assessed risk meets or exceeds the configured threshold. Describe and document run in parallel after the gate. The deploy job runs lint (with apply-patch: true to auto-commit fixes), then push, then create-pr.

Custom Steps — Write your own reusable steps
Workflows — How steps fit into workflows
Artifacts — What each step produces

On this page