release

Release

Contract

Prereqs:

• Run from the target repository root unless the user explicitly names another repository path.
• agent-run is installed from nils-cli 0.20.0 or newer and available on PATH.
• The target repository owns an executable .agents/scripts/release.sh.
• Versioning, changelog, tags, publishing, and post-release checks are owned by the consuming repository.

Inputs:

• User-provided release arguments, passed through to the project-local script.
• Optional repository path when releasing from another checkout.

Outputs:

• The project-local script's stdout, stderr, exit code, release URL, and verification evidence.
• A clear stop message when no project-local implementation exists.

Failure modes:

• .agents/scripts/release.sh is missing or is not executable.
• agent-run is unavailable or reports a blocked required project environment.
• The project-local script exits non-zero.
• The requested repository path is not a directory.

Entrypoint

Resolve the repository root, then invoke the project-local script through agent-run so repository .envrc / .env decisions are explicit:

agent-run exec --cwd "$repo_root" -- ./.agents/scripts/release.sh "$@"

When the script is missing or not executable, report:

no project-local implementation: .agents/scripts/release.sh

Workflow

1. Resolve the target repository root.
2. Verify .agents/scripts/release.sh exists and is executable.
3. Run the script through agent-run exec --cwd "$repo_root" --, passing through user arguments.
4. Report the release evidence and exit code printed by the script.
5. Do not run generic release commands when the project-local script is absent.

Boundary

Runtime-kit owns only the dispatch contract. Each consuming repository owns its release implementation and approval gates.