tapd-story-pipeline

Invoke the pipeline with the story ID and workspace ID, for example by saying implement story #1234567890 in workspace 20000001. The pipeline pulls the requirement via TAPD MCP, runs specify, plan, and tasks stages, then pauses at a confirm checkpoint for your review before implementing and committing.

It runs six stages in order: specify (technical clarification and spec.md), plan (development plan and research), tasks (task breakdown and analysis), implement (TDD coding), validate (architecture, security, code review, and test coverage checks), and commit (statistics, commit message, git commit, TAPD status update).

Re-invoke the pipeline with a semantic command matching the checkpoint: approve or reject for confirm checkpoints, answer after filling questions.md for blocked checkpoints, retry after fixing content files for fail checkpoints, or abort to stop. The pipeline reads meta.yaml and continues from the recorded phase.

Yes. When the specify stage classifies the story as fullstack, the implement stage runs two serial rounds: backend-developer first, then frontend-developer using the plan.md API contract as the authoritative interface. Validation fixes are routed to the appropriate developer agent by affected file paths.

If all findings are attributed to the code itself, an in-place fix subagent runs and the four validation checks repeat, up to three rounds per attempt. If findings trace back to insufficient spec or plan, the pipeline rolls back to the upstream phase while preserving the code and records unresolved findings for differential fixing.

Zero cost totals mean the PostToolUse hook is not installed in the host IDE settings, so cost-events.jsonl is never written. Merge the hook configuration from the skill's scripts/settings.json into the project's .claude/settings.json; the pipeline still works normally without it, only cost tracking is affected.