Dry-run of release_start: exactly what would be created — branches, MRs, predicted tags (computed from the tags that exist right now), file edits with resolved values, and pipeline variables. Strictly read-only; nothing is touched.

ALWAYS call this before release_start with the same arguments and show the user the plan; only call release_start after they confirm.

Start a release: build run state from the manifest and fire every ready node.

inputs are the manifest-declared release inputs (e.g. library versions). Returns immediately; pipelines run in GitLab on their own time. Call release_plan first and show the user what will happen.

Read-only view of the release DAG: node states, tags, pipeline URLs, gates.

One idempotent tick: read actual GitLab state and advance whatever is ready.

Safe to call any time; does nothing if pipelines are still running.

Clear a manual gate (e.g. 'qa-signoff') and advance the nodes it was blocking.

After a fix landed on a node's release branch: cut the next iterated tag and build it.

Explain why a node's pipeline failed (advisory; reads the failed job's log).

The release report (markdown): inputs, per-node tags, pipelines, produced versions, and values captured from job logs (e.g. Sonar report URLs).

With publish=true the report is also committed to the state repo (reports/.md).

Every release known to the state store, newest first, with a progress summary.

Use this to discover coordinates when the user doesn't name one ("what's in flight?", "what was our last release?").

Validate a release manifest YAML without touching anything.

Returns 'valid' with a topology summary, or the exact validation error. Use before committing manifest changes to the state repo.

A node's repository tags, newest first, read LIVE from GitLab — no run state needed.

Use this to answer "what was the last released version of ?" from the repo itself rather than from a release's state or the manifest.

The release manifest YAML — the declarative source of truth for the repo topology.