Dedupe Judge — Development agent for Claude Code
Decides whether pairs of blind-review findings describe the same defect.
How to install Dedupe Judge
Installs to ~/.claude/agents/wendellmcs-sworn-dedupe-judge.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/wendellmcs/sworn/HEAD/agents/dedupe-judge.md -o ~/.claude/agents/wendellmcs-sworn-dedupe-judge.md Restart Claude Code, or start a new session, for it to be picked up.
What Dedupe Judge does
name: dedupe-judge description: Decides whether pairs of blind-review findings describe the same defect. Only warns, does not change the count. Use inside /sworn:review, Step 3, never alone. model: haiku effort: low maxTurns: 3 tools: Read, Grep, Glob omitClaudeMd: true
You are the "same defect" judge. You receive, in the briefing, a **list of pairs** of findings from a blind review. Each pair has a `novo` side and an `outro` side, and each side carries `chave`, `ronda`, `lente`, `sever
Alternatives in Development
- Svelte Recurrence Sweep — Given a defect just found or fixed in a SvelteKit app (apps/moderator, apps/auth, apps/creator-studio), finds 7.2k ★
- Command Helper — Smart command assistant that understands user intent and suggests relevant commands 587 ★
- Mathodology Award Judge — Use for one independent blind judge seat scoring an award-level submission against contest rubric criteria 137 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Judge Router
Decides whether a stage's verdict requires cross-vendor or same-vendor judging by calling pp_harness.gate_elig
Vouch Judge
Isolated grounding judge for Vouch. Given a verification packet file (judge instructions, corpus evidence, num
Skill Eval Reporter
Compares repeated paired execution results using blind A/B methodology and generates a skill effectiveness rep
Disc Blind Reader
A blind reader of the stage-1 discovery review — reads the PR-FAQ and User Stories alone and commits to the co
Defect Skeptic
Use when one claimed defect in a codebase needs an independent, blind check before anyone acts on it, for exam
Brainswarm Judge
brainswarm judge role. Compares anonymised finalist pairs against the frozen rubric and rules on disputed crit
Related Skills
Newlife
Turn a curiosity into a question that can be settled, freeze what would count as an answer before running anyt
Escape
Analyze a defect that escaped to production — which layer went blind, and what changes: $ARGUMENTS
Editscore Ladder Run
Run an EditScore-7B pairwise ladder for an image-edit generator (SDEdit / diffusion / LoRA / distilled step-co