bmad-dev-auto

Invoke the skill with an intent such as a story ID, ticket number, or short feature description. It routes through clarify, plan, implement, and review steps automatically, halting only with an explicit terminal status like done or blocked.

Step 4 launches two parallel review subagents, a Blind Hunter and an Edge Case Hunter, on the diff since the baseline revision. Findings are deduplicated, severity-rated, and triaged into intent_gap, bad_spec, patch, defer, or reject categories.

Patch findings are auto-fixed, bad_spec findings trigger a spec amendment and full re-implementation loopback capped at 5 iterations, and intent_gap findings halt the run as blocked. Deferred items are logged to a deferred-work file.

Yes. If the invocation points to an existing spec, its frontmatter status routes the run: draft goes to planning, ready-for-dev or in-progress to implementation, in-review to review, and done triggers a fresh follow-up review pass.

The workflow cannot ask questions, so unclear intent, dirty working trees, missing previous-story continuity, or intent gaps cause a blocked halt. Subagents must run synchronously since there is no event loop to resume backgrounded tasks.