Skip to content

hashgraph-online/awesome-codex-plugins / plugins/boshu2/agentops/skills/council

council

Verified

Compare independent opinions from models or contexts without inflating agreement. Use when: wanting several opinions or a debate, or summarizing several reviewers' results.

FileKindSize
SKILL.mdskill md12.2 KB
references/debate.mdmarkdown3.8 KB
references/duel.mdmarkdown1.2 KB
references/interview-panel.mdmarkdown1.4 KB
references/judge-split.mdmarkdown1.4 KB
schemas/council-report.v1.schema.jsonconfig4.7 KB
scripts/validate-output.shexecutablescript4.3 KB
scripts/validate.shexecutablescript4.6 KB

references/debate.md

View on GitHub
# Debate and majority selection

Loaded by [Council](../SKILL.md) when the caller selects a bounded debate or a
voting rule. Independent comparison of sealed initial views needs neither.

## Rounds

Every round uses fresh contexts with new observed IDs, distinct from the author,
synthesizer, and prior rounds. Initial participants must not see peer answers or
the author's preferred conclusion. Seal all initial responses before sharing
any. Reused or colliding IDs stop reliance on that round: repair the isolation
within remaining bounds or disclose it as non-independent.

For debate, synthesize a candidate from sealed proposals and later objections;
the synthesizer does not vote. Share the same prior responses, evidence, and
exact candidate with every participant in the next round. Require substantive
challenges to competing claims, evidence for changed positions, and remaining
objections. Do not share partial current-round responses with peers. Fresh
contexts that receive earlier answers are **peer-informed deliberation**, not
new independent confirmations; label them separately from the initial views.

Before the next round, check each disputed factual claim that could change the
candidate at its cited source within the authorized evidence, and mark it
verified, contradicted or unverifiable with the location read. Every
participant receives the same marked claims; a contradicted claim stays visible
beside its contradiction, never quietly removed. Marking a fact is not a vote.

Before debate, fix the maximum rounds and total deadline from the caller/native
bounds; clarify missing bounds before launching. Initial independent proposals
are round zero, outside the debate-round count. New contexts, revisions, and
retries never renew the deadline or round allowance. Stop at the agreed
condition or exhausted bound and report unresolved disagreement honestly.

## Majority selection

If the caller requests majority selection, record the fixed participant roster,
threshold, and whether distinct models or judges are counted. A majority means
more than half of that fixed denominator; count each selected model once for a
model majority. Each participant returns support, oppose, or abstain for the
**same exact candidate digest**; only unconditional support counts. Required
amendments mean oppose, not support for a private revision. A changed candidate
requires a new digest and fresh round; never carry old votes forward. Do not
shrink the denominator for missing responses, errors, or abstentions, and never
replace a required model with an available one. All caller-required legs must
return eligible views before claiming the requested council is complete. Stop
once a completed round meets the selected threshold; otherwise return no agreed
recommendation at the cap.

Report the tally as **deliberative agreement** and retain minority objections,
even when unanimous. A majority can select an advisory design recommendation;
it cannot establish factual truth, measured benefit, validation acceptance, or
resolve a failed required validation leg. Without a caller-selected voting
rule, synthesize the evidence without inventing a vote.

Record round, mode and candidate digest in each `judgment`, and bounds, roster,
threshold, tally and stop reason in `synthesis.notes`; keep initial and
deliberative support distinguishable.

## Example request

```text
Use /agentops:council to compare architectures for reliable Job redelivery.
Use four distinct available models I authorize for this source. Have each
propose an approach independently, then debate the alternatives. Require
three of four to support the same exact recommendation. Cap debate at five
rounds and the whole council at 60 minutes. Preserve objections and explain
what evidence we still need before implementation or validation.
```

Resolve the authorized model pins before dispatch; these bounds are the
caller's choices, not skill defaults.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 322kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 322kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 322kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 322kUpdated Claude CodeCodexVerifiedInstall