a5c-ai/Babysitter
Run babysitter-orchestrated AI workflows via agent-platform yolo. Harness-agnostic: supports pi, claude-code, codex, gemini-cli, and more.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Jul 3, 2026
- License
- MIT
Pinned Snippet
uses: a5c-ai/babysitter@b10d119588538f19f0596a157f12218256b17aaa # v0.0.188tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| prompt | Task prompt for babysitter. GitHub context environment variables are available for shell expansion ($GITHUB_REPOSITORY, $GITHUB_EVENT_NAME, $GITHUB_SHA, etc.). If omitted, babysitter derives the task from the triggering event context. | no | — |
| harness | AI harness to use for worker tasks. Options: internal (default, programmatic pi agent), pi, claude-code, codex, gemini-cli, opencode, github-copilot, cursor. | no | internal |
| model | Model identifier in provider:modelId format (e.g. anthropic:claude-sonnet-4-20250514, openai:gpt-4.1, azure-openai-responses:gpt-4.1) or bare modelId (e.g. claude-sonnet-4-20250514). Provider is resolved from the model string; only the matching API key is required. | no | — |
| process-path | Path to an existing babysitter process definition file (JS). When provided, Phase 1 (interview/process creation) is skipped and the process is used directly. | no | — |
| workspace | Working directory for the harness. Defaults to the repository checkout path. | no | ${{ github.workspace }} |
| max-iterations | Maximum orchestration iterations before the run terminates. | no | 256 |
| runs-dir | Directory for babysitter run state (journals, tasks, blobs). | no | .a5c/runs |
| timeout-minutes | Timeout guidance in minutes. Note: composite action steps cannot use dynamic timeout-minutes. Set timeout-minutes on your workflow job instead (e.g., jobs.<id>.timeout-minutes: 30). This input is provided for documentation purposes and can be read by custom wrapper workflows. | no | 30 |
| verbose | Enable verbose debug output. | no | false |
| anthropic-api-key | Anthropic API key. Required for claude-code harness or Anthropic models via internal harness. | no | — |
| openai-api-key | OpenAI API key. Required for codex harness or OpenAI models via internal harness. | no | — |
| gemini-api-key | Google Gemini API key. Required for gemini-cli harness or Gemini models via internal harness. | no | — |
| azure-openai-api-key | Azure OpenAI API key. Required for Azure-hosted models. | no | — |
| azure-openai-project-name | Azure OpenAI resource/project name (maps to AZURE_OPENAI_PROJECT_NAME). | no | — |
| azure-openai-deployment | Azure OpenAI deployment name (maps to AZURE_OPENAI_DEPLOYMENT). | no | — |
| azure-openai-base-url | Azure OpenAI base URL override (maps to AZURE_OPENAI_BASE_URL). | no | — |
| github-token | GitHub token for API access. Defaults to the automatic GITHUB_TOKEN. | no | ${{ github.token }} |
Outputs
| name | description |
|---|---|
| run-id | The babysitter run ID. |
| run-dir | Path to the babysitter run directory. |
| status | Final run status: completed, failed, or timeout. |
| iterations | Number of orchestration iterations executed. |