Parallel Run

Parallel Run Skill

A user invocation (/parallel-run <parallel-slug>) forks the parallel-orchestrator agent to execute a parallel run that parallel-planner has already prepared. The parallel slug (or a path under docs/features/parallel/) for this run is:

$ARGUMENTS

Procedure

1. Resolve the parallel home. A bare slug resolves to docs/features/parallel/<parallel-slug>/; a path argument resolves to its containing parallel run folder.
2. Resolve the committed kickoff artifact docs/features/parallel/<parallel-slug>/parallel-kickoff.md. Discovery is a single local path lookup in the invoking worktree. This surface has no integration branch, so there is no integration ref to fetch, no ref-existence probe to run, and no ref-reading fallback to attempt: the durable copy of the kickoff artifact is committed by /parallel-plan on the planner branch parallel/<parallel-slug>-plan and is reachable from the checkout that invokes this skill.
   • STOP without delegating anything when that path does not exist. Report that the run has no committed kickoff artifact: the user must run /parallel-plan first (or, for a run whose manifest was authored manually, invoke /parallel-orchestrate <manifest-path> directly).
   • Do not synthesize a kickoff prompt, do not fall back to the planner's working copy under artifacts/orchestration/, and do not launch a partial run in place of the STOP.
3. Execute the kickoff artifact's ## Invocation Prompt section as the run objective, applying the parallel-orchestrate skill procedure and the ## Prepared-Run Execution section of .claude/agents/parallel-orchestrator.md: each item's child Agent(orchestrator) delegation resumes at atomic execution from that item's committed plan-path rather than re-running promotion, research, or planning. Feature-document authoring and preflight clearance are likewise not repeated: those outputs are already committed and preflight-clear on each item's own pushed feature branch.
4. Honor existing checkpoint state: if artifacts/orchestration/parallel-orchestrator-state.json already tracks this parallel_slug, resume per the parallel-orchestrate skill's resume procedure instead of restarting, re-deriving durable ground truth from git worktree list --porcelain, git branch, and gh pr view --json state,mergedAt,headRefOid before acting on any recorded value.

Scope

This skill adds no procedure of its own beyond kickoff-artifact resolution. Cohort consumption and ordering, the cohort barrier, max_concurrency slot filling, the per-item branch and worktree lifecycle, the child kickoff parameter, model selection, per-item merge to main on green, per-item merge-conflict handling, worktree cleanup, parallel-status.md maintenance, checkpoint persistence, and mode-dependent completion are governed entirely by the parallel-orchestrate skill (.claude/skills/parallel-orchestrate/SKILL.md).

There is no final integration pull request on this surface, and no fan-in step follows the items: each item's pull request targets main directly and is merged individually by parallel-orchestrator after its checks are durably confirmed green.