Runtimes

A runtime is the CI/CD environment where autoducks agents execute. The runtime is responsible for:

• Listening for triggers (issue comments, PR merges, workflow dispatch)
• Setting environment variables for the agent scripts
• Granting repository permissions to the agent process
• Invoking pre.sh, the LLM agent, and post.sh in sequence

The agent scripts themselves (pre.sh, post.sh, and the core utilities in .autoducks/) are runtime-agnostic. Only the workflow definitions that wire triggers to scripts are runtime-specific.

Supported runtimes

Runtime	Status	Workflow files
github-actions	✅ Supported	.autoducks/runtimes/github-actions/

How runtimes work

Runtime workflow files live under .autoducks/runtimes/{runtime}/. At install time, the installer copies them to wherever the CI/CD platform expects to find them.

For GitHub Actions:

.autoducks/runtimes/github-actions/autoducks-*.yml
    → .github/workflows/autoducks-*.yml

The setup.sh script verifies that the deployed workflow files stay in sync with the canonical definitions in .autoducks/runtimes/.

Adding a new runtime

To support a new CI/CD platform, create a new directory under .autoducks/runtimes/{runtime}/ with workflow files that:

1. Listen for the appropriate triggers (issue comment, PR merge, workflow/pipeline dispatch)
2. Set the required environment variables (ISSUE_NUM, BASE_BRANCH, REPO, COMMENT_ID, RUN_ID, GH_TOKEN or equivalent)
3. Call the agent pre.sh, then the LLM step, then post.sh — or run.sh for bash-only agents

The agent scripts and provider implementations do not need to change.

Note

The trigger surface is the main integration point between runtimes and the rest of autoducks. As long as the correct environment variables are set, any CI/CD platform that can run bash scripts can host an autoducks runtime.