runlisted

# Dev Pipeline Fully autonomous pipeline: GitHub issue → branch → implement → review → PR. **Runtime:** Claude Code CLI with `--permission-mode auto`. Not designed for GitHub Actions — the pipeline depends on Claude Code's Agent tool for intake, planning, implementation, and code review. **How to run:** One default mode plus a diagnostic escape hatch. The pipeline runs interactively in a Claude Code session (subscription-covered per ADR-015); the interactive escape hatch is for manual recovery and first-use priming, not a normal way to run the pipeline. **Default — `/dev-pipeline:run <issue-number>`** in your existing Claude Code session. Runs all stages (1–10) in this one session with no prompts at the gates (autonomous fail-fast contract — see "Skill-body failure contract" below) and ends with a draft PR URL. Stage 8's reviewer fan-out runs as fresh `agent()` calls inside the `Workflow` script (`workflows/code-review.mjs`), so each reviewer gets clean context; synthesis runs in-session on the caller's model. **Diagnostic fallback — `DEV_PIPELINE_MODE=interactive /dev-pipeline:run <issue-number>`.** Not a peer mode; a manual-recovery / first-use-priming escape hatch. Same single-session in-process flow as the default, but the per-stage gates **prompt** for a decision instead of taking the autonomous fail-fast write. Use only when an autonomous abort needs to be debugged interactively — e.g., stepping through after a `failureContext.reason` was written. ```bash # Batch