Menu

Individual skill

wise-men-flash

Use when the user asks for a "flash council", "quick council", "fast panel", "wise men flash", "quick second opinion from a panel", or invokes /wise-men-flash — and when they want a council-quality answer to a hard question fast (four subagent calls and about a third of the full council's time, a median 16.7 minutes in its own pre-registered round) rather than the full wise-men protocol. A three-member council of a practitioner anchor named for the job the question belongs to, one domain member, and a Devil's Advocate — each reasoning down a different path, spawned in parallel. No peer-review round, no debate; the main thread synthesizes one decision memo, and a single fresh checker verifies it before the user sees it. Members are offline (Read/Grep/Glob only). Its answer is a decision memo — Recommendation / Why / What to do / Risks / Strongest counter-position / Confidence — with no council mechanics in it. Sibling of the full `wise-men` skill, with the same answer format and one fixed fast tier instead of five. Pure Claude — no external APIs.

SKILL.md

Read pinned source

Source commit b8db7ddada08275fb16c25e526a84823e62a85fa

What was checked

Matches the observed repository HEAD
Yes
Scanner rules are current
No
Repository scan complete
No
Skill scan complete
No
Observed behavior
Not checked yet
Last checked
Nov 4, 2026

These are observations at the recorded commit, not a guarantee of safety. Review the source before running it.

Install this skill

Choose an agent in your Stack or copy a command below. Commands fetch the pinned commit and copy only this skill directory. Existing destinations are refused. Copying a command does not install anything.

An install command is unavailable: current rules, complete repository and skill scans, an eligible behavior level and a permitted license are required.

Repository context

Repository stars (not skill ratings)
10
Repository last updated
Nov 4, 2026
License
MIT

Observed agent integration: claude-code, codex

Files and excerpt

File metadata and permitted, redacted excerpts come from the recorded scan. Source links are pinned to that commit.

Browse files
Read excerpt

# wise-men-flash

The fast tier of the wise-men council, extracted into its own skill. Three members, no peer-review round, one synthesis check. It answers in four subagent calls and about a third of the full council's time. Measured as installed in the pre-registered round-5 head-to-head (eight new questions, nine arms, three blind judges each): 23.4/25, clearly ahead of all seven rivals — six other skills and a plain answer — on every question, and a quarter point behind the full `wise-men` council (23.6) with an interval that includes zero, in a median 16.7 minutes and 3.95 USD a question against its 50.2 minutes and 17.72 USD (list price). Round 4 had measured the same profile as the wise-men fast arm: 23.3/25 in 10.8 minutes. Use it when the decision is real but does not justify the full protocol, or when subagent budget is tight.

**When to reach for the full `wise-men` instead**: an irreversible or high-blast-radius call, a compound question that needs splitting, or anything where you want peer-review scoring and a debate round. Flash is one speed. It does not adapt its depth to the stakes — that is exactly what `wise-men` (solo → quick → standard → deep → paranoid) is for. When in doubt on a genuinely high-stakes, hard-to-reverse decision, spend the full council.

## Orchestrator runbook (execute this; the reference sections below explain it)

You (main thread) are the orchestrator and the Chairman. These steps are MECHANICAL. Every deviation MUST be disclosed in the final output's footer — a silently shortcut council is the fake-council anti-pattern.

0. **Pre-flight** (4 checks). If the user already invoked the skill, do NOT ask "want a council?" — they asked for one. If the question is trivial, one-line, a settled fact, or a choice reversible in minutes, say so and answer directly; a council is overkill. If it is compound, split it and say which part you are answering. If it hinges on a current external fact (a price, a version, a tool's present state), gather that fact ONCE now and put it in the brief with its date — never let offline members answer time-sensitive questions from training data.
1. **Roster** (Stage 0): three members.
   - **Practitioner anchor** — named for the actual job the question belongs to (an SRE lead, an employment lawyer, a pricing lead, a launch DevRel lead). Owns correctness and completeness: the concrete first step and the must-check facts. **Strong model.**
   - **One domain member** — the sharpest single lens for this question's domain (see the heuristics below). Owns the non-obvious frame. **Mid model.**