Snapshot tests for AI agent tool-call traces. Wraps the npm-published @mukundakatta/agentsnap library.

The action diffs each *.snap.json baseline in your snapshots dir against a sibling *.current.json (produced by your test run earlier in the workflow via record()) and fails the PR on drift. On drift, posts a Markdown summary as a PR comment.

- uses: actions/checkout@v4
- run: npm test          # your tests must record a *.current.json next to each *.snap.json
- uses: MukundaKatta/agentsnap-action@v1

1. Your test calls record(fn) from @mukundakatta/agentsnap and writes the trace to tests/__agentsnap__/<name>.current.json.
2. This action looks for tests/__agentsnap__/<name>.snap.json (the baseline) and runs diff().
3. If the diff status is in fail-on-drift, the action fails the workflow.
4. If comment-on-pr is true and the event is a PR, the action posts a drift summary as a PR comment.
5. Set update-snapshots: true to overwrite baselines instead of failing — useful for the "I changed the prompt on purpose" case.

To post PR comments, the workflow needs:

permissions:
  pull-requests: write
  contents: read

name: Agent regression
on: pull_request

permissions:
  pull-requests: write
  contents: read

jobs:
  agentsnap:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20 }
      - run: npm ci
      - run: npm test    # writes tests/__agentsnap__/*.current.json
      - uses: MukundaKatta/agentsnap-action@v1
        with:
          snapshots-dir: tests/__agentsnap__
          fail-on-drift: REGRESSION,TOOLS_CHANGED

Part of the @mukundakatta agent stack:

• agentvet-action — lint LLM tool definitions
• mcp-stack-validate-action — single CI gate that runs the whole stack

MIT