Eval Case Auditor — Development agent for Claude Code
Read-only auditor for triage eval cases.
How to install Eval Case Auditor
Installs to ~/.claude/agents/anthonycoffey-support-triage-agent-eval-case-auditor.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/anthonycoffey/support-triage-agent/HEAD/.claude/agents/eval-case-auditor.md -o ~/.claude/agents/anthonycoffey-support-triage-agent-eval-case-auditor.md Restart Claude Code, or start a new session, for it to be picked up.
What Eval Case Auditor does
name: eval-case-auditor description: Read-only auditor for triage eval cases. Use after adding several cases or before promoting a baseline. Reports ambiguous labels, leaked answers, shortcuts and coverage gaps. Never edits files. tools: Read, Grep, Glob
You audit evaluation cases for a support-ticket triage agent. You do not edit anything. You report findings.
Inputs to read:
src/agent/prompt.tsholds the policy that gold labels must follow.src/data/kb.tsand `src/data/custom
Alternatives in Development
- Ontology Term Lookup — Use this agent when you need to find ontology terms by their textual labels or descriptions using the OLS4 MCP 163 ★
- Learning Reviewer — Internal agent for reviewing learnings and promoting them to rules or skills 150 ★
- Eval Implementer — Implementer role of the D365FO agent eval loop 136 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Label Classifier
Assigns type:, priority:, effort: labels to a backlog item, each with one-line reasoning. Returns 'unclear' pe
Bk Researcher
Answers a technical question a skill cannot verify locally, from independent sources and the pinned major's do
Skill Eval Runner
스킬 eval 용 실행 러너 — 테스트 프롬프트 1건을 통제 조건에서 완료하고 원본 결과를 반환한다. skill-creator 스킬이 baseline vs 후보 대조 실행에 사용.
Baseline Tester
Runs one eval prompt with NO skill, as a plain Claude session would answer it. Use for the baseline condition
Dnp TDD Developer Hard
🔬 Deep TDD for complex .NET tasks: architectural decisions, ambiguous edge cases, high-risk refactoring. Writ
Prosecutor
Reviews a change through one lens and argues the case against it, filing findings that each carry a quote of t
Related Skills
Loki MCP Server
Read-only MCP server for Grafana Loki: lets AI agents (Claude Code, Cursor, Copilot) discover labels, read bou
Add i18n Message Key
Use when adding or changing visible copy, metadata, labels, validation messages, alt text, aria text, toasts,
Toolward
Security auditor for AI agent extensions — MCP servers, skills, plugins, connectors. Finds prompt injection, t