Skip to content

ValentinNikolaev/agent-skills / agent-plugins/skills/council

council

Verified

Run independent advisor passes and anonymized peer review for a consequential choice. Use when the user requests a council, debate, or stress test; prefer specialist reviews otherwise.

FileKindSize
SKILL.mdskill md7.0 KB
agents/openai.yamlconfig218 B
references/advisors-and-prompts.mdmarkdown4.4 KB
references/example.mdmarkdown2.5 KB

references/advisors-and-prompts.md

View on GitHub
# Council advisors and prompt contracts

Use this reference after selecting quick or full mode.

## Contents

- [Advisor styles](#advisor-styles)
- [Selecting quick-mode advisors](#selecting-quick-mode-advisors)
- [Advisor prompt](#advisor-prompt)
- [Reviewer prompt](#reviewer-prompt)
- [Synthesis prompt](#synthesis-prompt)
- [Independence rules](#independence-rules)

## Advisor styles

### Contrarian

Search for failure modes, missing constraints, downside, and reasons the proposed framing may be wrong. Distinguish a plausible risk from a sourced fact. Name the evidence needed to confirm the concern.

### First-Principles Thinker

Identify the underlying objective, strip away inherited assumptions, and test whether the stated options solve the actual problem. Propose a better framing when warranted.

### Expansionist

Look for underweighted upside, adjacent opportunities, option value, and asymmetric gains. Do not ignore cost or claim upside as fact; state the conditions required.

### Outsider

Use only the supplied frame. Surface jargon, hidden knowledge, stakeholder confusion, and assumptions an informed insider may miss. Do not invent domain context.

### Executor

Test feasibility, sequencing, reversibility, resource constraints, and the fastest credible validation step. Prefer actions that produce decision-relevant evidence.

## Selecting quick-mode advisors

Use three styles that create the strongest tension for the decision. Default to Contrarian, First-Principles Thinker, and Executor. Substitute:

- Expansionist when upside or option value is central;
- Outsider when audience comprehension, onboarding, or expert blind spots dominate.

Record the selected styles in the synthesis. Do not imply that omitted styles agreed.

## Advisor prompt

```text
Act as the [ADVISOR STYLE] in a decision council.

Thinking contract:
[ADVISOR DESCRIPTION]

Decision frame:
---
[SHARED FRAME]
---

Analyze independently. Do not assume facts that are absent from the frame.
Separate evidence, inference, and unknowns. State:

1. your central assessment;
2. the strongest supporting reason;
3. the weakest point in your preferred option or reasoning, and why it matters;
4. the largest risk or opportunity from your angle;
5. the evidence that could change your view;
6. one proposed action.

Stay direct and concise. Preserve uncertainty instead of manufacturing certainty.
```

## Reviewer prompt

```text
Review anonymized decision-council responses against the shared decision frame.

Decision frame:
---
[SHARED FRAME]
---

Responses:
[ANONYMIZED RESPONSES]

Do not infer advisor identities. Evaluate reasoning rather than rhetoric or majority count.
Return:

1. the best-supported response and why;
2. the largest unsupported leap or blind spot;
3. the weakest point that remains in the leading option or argument, even if it passes;
4. what every response missed;
5. the evidence most likely to change the decision;
6. any safety or professional-review boundary.
```

## Synthesis prompt

```text
Synthesize the decision frame, advisor responses, peer reviews, and verified sources.

Do not treat model agreement as independent confirmation. Prefer evidence quality over votes.
Preserve material disagreement and calibrate uncertainty.

Use this structure:

## Where the Council Converges
[Shared reasoning; explicitly distinguish convergence from proof.]

## Where the Council Disagrees
[Competing premises and the evidence behind them.]

## Blind Spots and Evidence Gaps
[Missing facts, options, and professional-review needs.]

## Weakest Points
[Explain the weakest supported part of the leading option or argument, its
practical effect on the verdict, and whether it needs a change or is an
accepted limit. For scored work, explain the lowest-rated criterion.]

## Recommendation
[Clear recommendation, conditional where necessary.]

## First Action
[One concrete, reversible next step.]

## What Would Change the Verdict
[Decisive evidence or threshold.]
```

## Independence rules

- Give every advisor the same neutral frame.
- Do not show an advisor earlier outputs before its own response completes.
- Anonymize content before peer review; remove role-revealing headers where practical.
- Do not tell reviewers the expected winner or the synthesizer's preliminary view.
- Do not count repeated claims as corroboration unless they trace to independent evidence.
- When using one model for separated passes, disclose that the outputs are analytically separated but still correlated.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 323kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 323kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 323kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall