tapd-iteration-runner

Provide the iteration ID and workspace_id, and the runner executes four stages: environment init, dependency analysis, a per-story implementation loop, and a final report. Each story is implemented by the embedded tapd-story-pipeline until its phase reaches committed.

The analysis stage pulls all stories via TAPD MCP stories_get, identifies parent-child relations from parent_id, extracts explicit dependencies from each story's dependency section, and detects technical and business dependencies. It then builds a DAG and topologically sorts stories into an execution sequence.

Yes. The init stage detects an existing iteration-state.json and runs a recovery flow that shows progress, backfills derived phase and stats caches from each story's meta.yaml, checks consistency, and re-enters the scheduling loop at the first unfinished story.

The runner handles iteration-level orchestration across many stories, while tapd-story-pipeline implements a single story. Single-story requests should trigger the pipeline directly; the runner only activates for iteration-level batch work.

Yes. All git and file operations include both Bash and PowerShell variants. On Windows, set the console output encoding to UTF8 before the first run to avoid garbled characters.

The report stage generates a changelog.md from each story's commit.md and a summary.md with code change statistics plus three-layer cost metrics covering iteration totals, per-story breakdowns, and per-stage credit and token consumption.