council-reference
Scannednpx machina-cli add skill rube-de/cc-skills/council-reference --openclawCouncil Reference Data
Structured Response Format
Every council agent MUST return this JSON structure:
{
"consultant": "gemini|codex|qwen|glm|kimi|claude-deep-review|claude-codebase-context",
"success": true,
"fallback": false,
"confidence": 0.0-1.0,
"severity": "critical|high|medium|low|none",
"findings": [
{
"type": "security|performance|quality|architecture|bug|documentation",
"severity": "critical|high|medium|low",
"description": "...",
"location": "file:line",
"recommendation": "..."
}
],
"summary": "One-paragraph summary"
}
location: MANDATORY for /council review findings. Format: file:line (e.g. src/api.ts:42).
Expertise Weight Matrix
Task Gemini Codex Qwen GLM-5 Kimi K2.5
Security 0.90 0.80 0.70 0.75 0.70
PR Review 0.85 0.90 0.80 0.75 0.80
Architecture 0.85 0.70 0.65 0.80 0.75
Code Quality 0.70 0.80 0.90 0.70 0.80
Performance 0.75 0.85 0.85 0.70 0.80
Brainstorm 0.65 0.60 0.90 0.85 0.80
Algorithms 0.70 0.75 0.85 0.85 0.80
Debugging 0.75 0.90 0.80 0.75 0.80
Scoring Thresholds
Score Meaning
───── ───────────────────────────────────────────────────────
0 False positive. Doesn't hold up to scrutiny.
25 Might be real, but unverified. Could be false positive.
50 Real but minor. Unlikely to occur in practice.
75 Verified real. Will impact functionality. Important.
100 Confirmed. Frequent in practice. Evidence conclusive.
───── ───────────────────────────────────────────────────────
Filter threshold: Only findings scoring >= 80 appear in the final report.
False Positive Taxonomy
Score 0 (auto-reject) if a finding matches any of these:
- Pre-existing issue not introduced in the current changes
- Problem that a linter, typechecker, or compiler would catch
- Pedantic nitpick that a senior engineer would not call out
- General code quality complaint NOT backed by project CLAUDE.md
- Issue on lines that were NOT modified in the changes under review
- Intentional functionality change clearly related to the broader change
- Code with explicit lint-ignore or suppress comments
Synthesis Formula
Weighted Score = Σ(Opinion × Expertise × Confidence) / Σ(Expertise × Confidence)
Source
git clone https://github.com/rube-de/cc-skills/blob/main/plugins/council/skills/council-reference/SKILL.mdView on GitHub Overview
This skill enforces a standardized JSON payload for council reviews, including consultant identity, confidence, severity, findings, and a one-paragraph summary. It codifies the Expertise Weight Matrix, Scoring Thresholds, and False Positive Taxonomy to ensure consistent, evidence-based evaluations. The Synthesis Formula combines opinions, expertise, and confidence to derive a weighted score.
How This Skill Works
Council subagents emit a JSON document matching the required schema, including an array of findings with type, severity, description, location, and recommendation. Location is mandatory for /council review findings. The system applies the Expertise Weight Matrix and Scoring Thresholds to compute a final confidence-weighted score via the Synthesis Formula.
When to Use It
- Review code changes for security, performance, or architecture in a PR and report findings with file:line locations.
- Aggregate findings from multiple consultants (Gemini, Codex, Qwen, GLM-5, etc.) and compute a final score using the weight matrix.
- Filter to final report by the 80+ threshold to avoid false positives.
- Characterize findings with explicit types (security, performance, architecture, code quality, etc.) and a severity level.
- Use the false positive taxonomy to auto-reject non-actionable issues (pre-existing issues, lint issues, etc.).
Quick Start
- Step 1: Initialize a council review payload with a selected consultant and context for the changes under review.
- Step 2: Add findings including type, severity, description, location (file:line), and a concrete recommendation.
- Step 3: Run synthesis to compute the weighted score and ensure only findings with score >= 80 appear in the final report.
Best Practices
- Always include location (file:line) for council review findings.
- Populate findings with type, severity, description, location, and a clear recommendation.
- Validate the final report against the 80+ threshold to include only meaningful findings.
- Apply the false positive taxonomy to minimize noise from non-actionable items.
- Provide a concise one-paragraph summary in the summary field for quick executive understanding.
Example Use Cases
- PR audit reporting security and architecture findings across multiple files with explicit file:line locations.
- Multi-agent review using the Expertise Weight Matrix to identify top-scoring findings.
- Final report containing only findings scoring >= 80 with a clear, actionable recommendation.
- Findings described with type, severity, description, location, and recommended remediation.
- Report includes a summary field that encapsulates the overall risk and impact.
