Sigma Step Judge — Development agent for Claude Code
Fresh-context cross-family judge for Sigma step artifacts.
How to install Sigma Step Judge
Installs to ~/.claude/agents/dallionking-sigma-protocol-public-sigma-step-judge.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/Dallionking/sigma-protocol-public/HEAD/.claude/agents/sigma-step-judge.md -o ~/.claude/agents/dallionking-sigma-protocol-public-sigma-step-judge.md Restart Claude Code, or start a new session, for it to be picked up.
What Sigma Step Judge does
name: sigma-step-judge description: "Fresh-context cross-family judge for Sigma step artifacts. Runs via fusion: a cross-family panel analyzes the artifact against a rubric, then the synthesizer produces a structured PASS/FAIL verdict with per-criterion evidence quotes. Advisory-until-calibrated (bal-acc ≥0.85, TNR ≥0.80, n≥40, cross-family). Used as the gap-analysis exit gate and the C1c CRAFT semantic scorer." tools:
- Read
- Bash model: opus permissionMode: default skills:
- verific
Alternatives in Development
- Dashclaw Gate Runner — Runs the DashClaw verification gates (lint, full vitest suite, build, contract checks) and returns ONLY the fa 297 ★
- Uxaudit L3 Judge — L3-vision Judge for the uxaudit pipeline 52 ★
- Qrspi Finding Verifier — Score a single reviewer finding 0–100 against the /code-review confidence rubric 38 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Harness Checker
Fresh-context artifact evaluator for goal loops. Reads only final artifacts — never Maker reasoning or PROGRES
Prose Judge
Fresh-context fidelity judge for the prose bench. Reads ONE gate file (docs/prose-rubric.md, the materials, on
Claim Extractor.Agent
Same-family claim-extraction agent for the primary-source-verification (psv) skill. Runs in a fresh, independe
Doc Reviewer.Agent
Same-family (同源) adversarial doc reviewer for the cross-source-review skill. Runs in a fresh, independent cont
Smith Inspector
Schema and invariant validator for AgentSmith artifacts. Fail-closed. Invoke whenever a draft agent/skill/comm
Gate
統合された position(現時点の答え)を成功基準(rubric)に一項目ずつ客観照合し、論理の穴・未対応の反論・深さ不足・過大主張を批判的に点検して、VERDICT: PASS/FAIL と不足の構造化フィードバッ
Related Skills
Squad Review
Run only the reviewer agent against the current run's code and tests. Produces a PASS/FAIL verdict and routes
LLM Fusion
OpenAI and Anthropic compatible LLM proxy for Ollama Cloud — panel→judge→synth fusion + smart routing, with ta
Extraction Judge.Prompt
Extraction Judge — Fact Durability Rubric