Wezel GitHub Action

Runs one exact experiment assigned by Fiflok. The workflow checks out the assigned commit, then passes Fiflok's run-id and experiment-name dispatch inputs to this action. The action never claims arbitrary queue work and never dispatches another workflow.

Usage

name: Wezel assigned run
on:
  workflow_dispatch:
    inputs:
      run_id:
        description: Fiflok run ID
        required: true
      experiment_name:
        description: Experiment to run
        required: true
      commit_sha:
        description: Assigned commit
        required: true
      project_dir:
        description: Directory containing the Wezel project
        required: true

permissions:
  contents: read

jobs:
  run:
    runs-on: ubuntu-latest
    concurrency:
      group: wezel-run-${{ inputs.run_id }}
      cancel-in-progress: false
    steps:
      - uses: actions/checkout@v7
        with:
          ref: ${{ inputs.commit_sha }}
      - uses: wezel-build/gh-action@v1.0.1
        with:
          token: ${{ secrets.WEZEL_RUNNER_TOKEN }}
          run-id: ${{ inputs.run_id }}
          experiment-name: ${{ inputs.experiment_name }}
          project-dir: ${{ inputs.project_dir }}

The action marks the assigned run running, executes:

wezel experiment run EXPERIMENT --run-id ID --save yes --output-format json

It uploads a zstd-compressed tar containing only report.json and the optional attachments/ tree, then marks the run complete. Any setup, execution, packaging, upload, or completion error is reported as failed when possible.

The runner image must provide tar and zstd (both are available on GitHub's Ubuntu-hosted runners).

Linting experiments on pull requests

command: lint preserves the existing local-only validation mode. It checks forager schemas, summaries, and patch applicability without using a token or touching a run.

name: Wezel lint
on:
  pull_request:
    paths: [".wezel/**"]

jobs:
  lint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      - uses: wezel-build/gh-action@v1.0.1
        with:
          command: lint

Lint does not run wezel project tool sync, so it does not update wezel.lock.

Inputs

Input	Default	Description
command	run	run executes an assigned run; lint validates committed config.
token	—	Fiflok runner-scoped token. Required for run; unused by lint.
run-id	—	Exact Fiflok run ID. Required for run.
experiment-name	—	Exact assigned experiment. Required for run.
api-url	https://api.wezel.build	Fiflok API base URL.
project-dir	.	Directory containing .wezel/.
wezel-version	v0.1.5-pre-pre.20260916103109+5a021a6	Exact Wezel release to install. Each Action release pins the CLI version it was tested with.
github-token	${{ github.token }}	Used only to download Wezel releases.

Outputs

Output	Description
status	complete or failed.
run-id	Assigned run ID.

Run backlinks

The action supplies WEZEL_RUN_BACKLINK and WEZEL_RUN_BACKLINK_LABEL for the current Actions run. Set either environment variable explicitly to override the generated backlink.