Skip to content

github-action.md

Latest commit

 

History

History
85 lines (69 loc) · 2.51 KB

github-action.md

File metadata and controls

85 lines (69 loc) · 2.51 KB

GitHub Action

AgentFit can run in pull requests that change agent instruction files.

name: AgentFit

on:
  pull_request:
    paths:
      - AGENTS.md
      - CLAUDE.md
      - GEMINI.md
      - .cursor/rules/**
      - .github/copilot-instructions.md
      - .github/instructions/**

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true

jobs:
  agentfit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: actions/setup-node@v5
        with:
          node-version: 22
          package-manager-cache: false
      - run: corepack enable
      - run: pnpm install --frozen-lockfile
      - id: agentfit
        uses: kingkyylian/agentfit@v1
        with:
          version: 0.1.13
          adapter: dry-run
          run-tasks: true
          fail-below-score: 70
          task-count: 5
          timeout-seconds: 900
          budget-usd: 1
          format: markdown

Inputs

Input Default Description
version 0.1.13 AgentFit npm package version to run through npm exec.
adapter dry-run Evaluation adapter to run.
fail-below-score 70 Fail the job when the JSON report score is below this integer from 0 to 100.
task-count 5 Number of generated fitness tasks.
run-tasks false Execute generated tasks in isolated worktrees. Install dependencies first when this is true.
timeout-seconds 900 Maximum seconds for each task run.
budget-usd 1 Maximum adapter budget in USD.
format markdown Human report format.

The action generates the human report and JSON report from one agentfit eval invocation, so real-agent adapters do not run the same task batch twice.

Outputs

Output Description
score Final AgentFit score.
report-path Markdown report path.
json-path JSON report path.

PR Comment

For a complete workflow that creates or updates one AgentFit pull request comment, see pr-comment-workflow.md.

Before/After Compare

For PRs that update agent instructions, store a baseline report from the target branch and compare it with the PR report:

- name: Compare AgentFit reports
  run: |
    npx @kingkyylian/agentfit@latest compare before.json after.json \
      --format markdown \
      --output agentfit-compare.md \
      --fail-on-regression

Use the compare output as the PR comment body when you want reviewers to see whether an instruction change improved or regressed the repository.