skills.rest
Zhachory1
Zhachory1COMMUNITY
Explore all skills by this author →
Website𝕏@zhachory1GitHub

council

Convene 3-6 specialist personas to debate and review high-stakes decisions.

Updated Jul 8, 2026
Search Tags:#code review#personas#debate#multi-agent review#red-team#design doc#decision review
One-click install
npx skills add https://github.com/Zhachory1/autopraxis --skill council-zhachory1
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: council
Source: https://github.com/Zhachory1/autopraxis/tree/main/vendor/agent-fleet/skills/council
Command: npx skills add https://github.com/Zhachory1/autopraxis --skill council-zhachory1

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Single-perspective reviews of high-stakes decisions (model changes, experiment readouts, design docs, serving-path PRs, build-vs-buy choices) often miss blind spots, suffer from false consensus, or inflate severity. This Skill orchestrates a bounded multi-persona debate that surfaces dissent, stress-tests a solo baseline decision, and produces a decision-grade synthesis with named dissents. ## Core Features & Use Cases - Multi-persona review: Selects 3-6 specialist personas (red-team, mvp, occams-razor, ml-scientist, ab-critic, and more) by mode (--mode ship|research|domain|exec|minimal), operator-forced roster, or task-signal table, with mandatory overlap checks. - Bounded debate protocol: Runs a blind first round followed by critique-before-concede reflection rounds (default 2, hard cap 4), with a SUSPICIOUS-FLIP detector that flags capitulation under peer pressure. - Durable transcript and journal: Persists artifacts, round-tagged positions, and synthesis to a room log, then appends a journal row for later blinded-judge evaluation of whether the council beat the solo and lens baselines. - Use Case: Before shipping a risky architecture change, run a ship-mode council so red-team, mvp, and occams-razor attack scope and complexity while domain personas validate technical soundness, yielding a ranked issue list and a SHIP/BLOCK verdict. ## Quick Start Ask the agent to run a council review on your design doc or PR, for example: "Run /council on this architecture doc and tell me if it is safe to ship."

Frequently Asked Questions about council

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-persona review of a design doc or PR?▼

Invoke the council trigger (e.g. /council) with the artifact under review. The orchestrator selects 3-6 personas, runs a blind first round plus reflection rounds, and returns a synthesis with ranked issues, named dissents, and a verdict.

What review modes does the council support?▼

Five modes: ship (default, auto-includes red-team, mvp, occams-razor), research (coverage-aware persona selection), domain (specialized lenses first), exec (ceo/cto/vp-eng/product-pm focus), and minimal (exactly 3 personas for low-stakes reviews).

Can I force a specific set of reviewer personas?▼

Yes. Pass --personas a,b,c or name an exact roster of 3-6 personas, and the orchestrator honors it unless a named persona does not exist. The overlap check still runs and the roster is marked as operator-forced.

How many debate rounds does a council run?▼

The default target is 2 iterations, clamped to a hard cap of 4. A SUSPICIOUS-FLIP detection can trigger one extra round beyond the target, but never past the absolute cap of 4.

Why does the council require transcript and journal capture?▼

Durable room logs and journal rows prevent lost transcripts and enable later blinded-judge checks of whether the council beat the solo baseline. If capture fails, the run is explicitly declared incomplete rather than summarized from memory.

When should I not use a full council review?▼

Skip it for small reversible changes or low-stakes reviews where --mode minimal or a solo review suffices. Councils add cost and latency, and a unanimous BLOCK from the default-3 personas may reflect their shared adversarial bias rather than fatal flaws.

Related Skills

Explore More Software Engineering Skills→
OpenLAIROpenLAIR
1.1k

ds-intake-audit

Audits and trust-ranks existing research assets to select the next workflow anchor.

Official
Advanced
scanadyscanady

strategy-decision-council

Convene a four-voice decision council to surface dissent and tradeoffs.

Community
Advanced
bigeasyfreemanbigeasyfreeman
2

fix-bug

Orchestrate bug fix workflows with investigation, brief creation, verification, and council review.

Community
Advanced
SirTheOracleSirTheOracle

adversarial-lite

Generate adversarial plans for small tasks using isolated Codex subagents.

Official
Intermediate