GitHub Actions

Pickled accepts cell filters on the CLI, which lets a GitHub Actions matrix split a scenario into one job per cell. Each job sees only the cell it owns, which gives a clean per-cell status badge.

A minimal job

name: pickled

on:
  pull_request:
  push:
    branches: [main]

jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
      - uses: oven-sh/setup-bun@v2
      - run: bun install
      - run: bunx @pickled-dev/cli check . --fail-on error
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

That runs every scenario in pickled.yml in one job.

Split a matrix into separate jobs

When you want each cell to surface as its own check, fan out via the workflow matrix and pass cell filters to the CLI.

jobs:
  cell:
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        interface: [quick]
        source: [readme, llms]
        toolset: [none, web]
    steps:
      - uses: actions/checkout@v6
      - uses: oven-sh/setup-bun@v2
      - run: bun install
      - name: pickled ${{ matrix.interface }}/${{ matrix.source }}/${{ matrix.toolset }}
        run: |
          bunx @pickled-dev/cli check . \
            --interface ${{ matrix.interface }} \
            --source ${{ matrix.source }} \
            --toolset ${{ matrix.toolset }} \
            --fail-on error
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

That produces four jobs (one per cell), each named after the cell it runs. A PR check page shows them as four independent statuses.

Audit-only pre-flight

pickled audit runs no LLM calls and is cheap. Use it as a pre-flight gate before the agent jobs:

  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
      - uses: oven-sh/setup-bun@v2
      - run: bun install
      - run: bunx @pickled-dev/cli audit . --fail-on error

  cell:
    needs: audit
    runs-on: ubuntu-latest
    # ... matrix as above

That fails fast on registered-source/trap mismatches before spending tokens on agent runs.

Secrets

ANTHROPIC_API_KEY for Claude Code and Anthropic API targets.
Per-MCP-server secrets (e.g., CONTEXT7_API_KEY) referenced in pickled.yml via ${UPPER_SNAKE_CASE} expansion. Bun auto-loads .env locally; in Actions, set them as job-level env from ${{ secrets.* }}.

Threshold

Set a global threshold to make the run pass or fail on a single score:

threshold: 60

Without a threshold, the report prints Overall: N / 100 and the workflow always succeeds (it only fails on the --fail-on flag, which targets per-scenario verdicts).

A minimal job

Split a matrix into separate jobs

Audit-only pre-flight

Secrets

Threshold

On this page