orchestratelisted

Your job is to orchestrate the execution of a plan through the multi-agent development pipeline by spawning subagents (Agent tool) for each step. You control the flow, handle retries, and ensure each agent's output feeds correctly into the next. > **Maintainer note:** This lives in a **skill**, not an agent definition, on purpose: the orchestrator must run in the main session, because only the main session can spawn subagents (a subagent cannot spawn others). Do not convert it into an agent. ## Arguments This pipeline runs for the plan: **$ARGUMENTS** If no plan name was provided, list available plans from the `plans/` directory and ask the user to pick one. ## Pre-flight Checks Before starting: 1. Read `plans/<plan-name>/plan.md` 2. Verify the plan status is "approved" (not "draft") and run `.claude/skills/orchestrate/scripts/plan-lint.sh <plan-name>`. A draft, or any `FAIL` line, goes back to `/plan-work` — spawn nobody. Then `go run ./tools/kb pack --plan <plan-name> --role orchestrator` — the lessons earlier runs paid for; read them now, not after the stumble. 3. Check the **Work Type** field to determine which agents to run: - `daemon` → skip web agents; run the E2E steps only if the plan defines `E*` acceptance criteria (a daemon-only change can still be E2E-observable through the dashboard) - `web` → skip daemon agents - `full-stack` → run all agents 3a. Check the **E2E Scope** field: - `new-specs` → Step 1 authors tests; expected verdict `authored`