parallel-dag-execution

Use to audit a spec before planning, or a DAG plan before execution, by fanning out independent single-concern lenses in parallel and reconciling their findings into one verdict. Invoke at either gate of the pipeline — after a spec is written (gate 1) and after `writing-dag-plans` produces a plan (gate 2). Do NOT use it to review completed code against a spec; that is `dag-spec-reviewer` / `dag-quality-reviewer` during execution.

Use when you have a DAG plan authored by `writing-dag-plans` and want to execute it with continuous parallel subagent dispatch. Maintains a ready queue, dispatches multiple implementer subagents per tick, runs spec+quality review per task, auto-retries-once on BLOCKED with model upgrade, and halts only the failed branch's downstream lineage on hard failure (parallel branches keep going). Resumable — re-invoke on a partially-executed plan to pick up where it left off.

Use to mutate a DAG plan mid-flight — add tasks, remove tasks, modify task bodies or `files:` scope, or rewire `depends_on:` edges. Operates only on `pending` and `ready` tasks; refuses to touch `running`, `done`, `failed`, or `skipped` history. Pairs with `writing-dag-plans` (same validation rules) and `executing-dag-plans` (which reads fresh state on every tick, so updates between ticks are safe).

Use when you have a spec/requirements for a multi-task implementation and want to author a DAG-aware execution plan with explicit per-task `depends_on` and `files` declarations. Pairs with `executing-dag-plans` for continuous parallel subagent execution. Invoke after `superpowers:brainstorming` produces a spec; do NOT use this for single-task work or when serial execution via `superpowers:writing-plans` is sufficient. When a spec fans out into multiple independently-executed plans, author a superspec-charter first, then use this skill per child plan (see the fan-out checkpoint).