The Board

The Board — multi-archetype decision review

A standing advisory board of distinct expert archetypes that reviews a decision from several opposed viewpoints and returns one aggregated verdict: PROCEED, CONCERN, or BLOCK — with each member's reasoning, the key risks, and the conditions that would change the call.

The archetypes are composite lenses (The Frontier Scientist, The Ruthless Operator, The CFO, The Safety & Alignment Reviewer, The Contrarian, …). They do not impersonate any real person — that is deliberate, both for integrity and to keep the skill clean to use and share.

When to use

Invoke this skill when the task is a judgment call, not a lookup:

• "Should we build / ship / fund / kill X?"
• "Is this a good idea?" / "Give me a second opinion."
• "Review this plan / strategy / architecture / launch."
• "Red-team this." / "Run a pre-mortem." / "What could go wrong?"
• Any go/no-go where multiple perspectives matter.

Skip it for factual lookups, simple how-tos, or tasks with one obviously-correct answer.

How to run a consultation

The script is deterministic (routing + tallying). You are the board — you adopt each archetype's doctrine and produce its verdict. Three steps:

1. Pick the panel and read the doctrines

python3 scripts/board.py panel --decision  --topic ""

--decision is one of general | venture | technical | launch | risk | all (aliases like gtm, architecture, safety are accepted; unknown → general; all convenes every archetype). This prints each selected archetype's doctrine and focus questions, plus the verdict schema.

2. Produce one verdict per archetype

For each archetype in the panel, reason strictly from its doctrine and focus questions — stay in character, be specific to the decision, and avoid boilerplate. Emit one object per archetype:

{
  "member": "strategist",
  "verdict": "PROCEED | CONCERN | BLOCK",
  "confidence": 0.0,
  "reasoning": "one or two sentences, specific to the decision",
  "key_risk": "the single biggest risk this archetype sees (or empty)",
  "key_condition": "what would change this verdict (or empty)"
}

Collect them into a JSON array (write it to verdicts.json, or pipe it).

3. Aggregate into the board's verdict

python3 scripts/board.py aggregate --file verdicts.json

Aggregation rules (deterministic):

• Any BLOCK → overall BLOCK.
• Strict majority CONCERN-or-worse → overall CONCERN.
• Otherwise → PROCEED.
• Confidence = mean of member confidences. Key risks and conditions are surfaced

from the dissenting members.

Then present the board's verdict to the user: the headline call, the tally, each member's one-line take, and the key risks / conditions. Be honest — if the board says CONCERN or BLOCK, lead with that.

Decision types → panels

| --decision | Panel focus | |---|---| | general (default) | Balanced go/no-go: Strategist, Operator, Safety, CFO, Contrarian | | venture | New product/business: Frontier Scientist, Operator, Strategist, CFO, Product Truth-Teller | | technical | Approach/architecture: Research Skeptic, Systems Architect, Safety, Frontier Scientist, Contrarian | | launch | Go-to-market: Operator, Growth Lead, Customer Advocate, Strategist, Compliance Counsel | | risk | Safety/compliance: Safety Reviewer, Compliance Counsel, Contrarian, Systems Architect | | all | The full board (every archetype) |

python3 scripts/board.py roster lists every archetype.

Tips

• Match the panel to the question; use all for a high-stakes, irreversible call.
• Make the archetypes genuinely disagree — the value is in the tension, not consensus.
• For your own org, you can add archetypes to data/archetypes.json or new panels to

data/routing.json (still: archetypes, not real people).

---

The Board is an open-source skill by AOS-1 — an operating system for trustworthy, governed AI. Multi-perspective decision review is the same discipline AOS-1 applies to autonomous agents.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: pamiray-m
• Source: pamiray-m/the-board
• License: MIT
• Homepage: https://aos-1.com

Install and usage instructions live in the source repository linked above.