dag-council

name	dag-council
tier	core
runtime	on-demand
trigger	mention
description	Auto-authored expert council as a DAG: give it a goal + ground truth, the conductor authors N distinct personas, they produce candidates concurrently, a multi-lens judge selects and grafts a winner. Grounded tier for domain-correctness forks. Trigger: multiple options / which approach / best of n / council / 多个方案 / 哪种方案好 / 拿不准选哪个 / domain fork / grounded / anti-happy-path decision / dag-council. Skip: single clear solution (just do it) / you want to hand-write the lens spec yourself (/dag-fanout) / needs web evidence first (/dag-research feeds this).
metadata	{"source":"oh-my-dag","version":"2.0.0","methodology":"conductor-authored personas + diversity>volume + multi-judge + graft runners-up + grounded (market-first + domain-role persona + anti-happy-path axis)"}
disable-model-invocation	true

/dag-council — auto-authored expert council

For wide solution spaces: the conductor reads your goal + ground truth and authors the personas itself (real, distinct expert angles — not one prompt resampled N times). Candidates generate concurrently, a judge panel scores through multiple lenses, and the champion is synthesized with the runners-up's orthogonal highlights grafted in.

Usage

bun run dag-council <goal-and-groundtruth.json>
# JSON: { goal, groundTruth, conductorModel?, lensCount?, lensModel?, reasonModel? }

Stdout: first line = artifact path, then LENS CHAMPIONS / SYNTHESIS CANDIDATES / FINAL marker sections (stable format, safe to split downstream).

Grounded tier (domain-correctness forks)

When the decision is in a domain where being wrong is expensive and reality is messy (accounting/legal/ops: dirty data, partial failure, lifecycle-end) and it's hard to reverse, upgrade the run:

1. Market-first: research what real-world practice actually does first (e.g. via /dag-research) and put that hard-evidence baseline into groundTruth — personas argue from facts, not vibes.
2. Domain-role personas: put the real roles into goal framing — the daily operator / compliance & audit / automation first-principles / lifecycle-end.
3. Anti-happy-path judging axis: state in groundTruth that every option must be judged against dirty data, concurrency, partial failure, cross-boundary, lifecycle-end and scale-balloon — otherwise personas quietly answer for the happy path.
4. Judge + graft: unanimous consensus is a strong signal; the champion often shrinks the debate rather than picking a side. Domain red lines stay with the owner — council output is input, not the final call.