Auditor Agent — Development agent for Claude Code
Self-improving auditor.
How to install Auditor Agent
Installs to ~/.claude/agents/yarkn24-maskpersona-ai-marcus-aurelius-auditor_agent.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/yarkn24/maskpersona-ai-marcus-aurelius/HEAD/agents/auditor_agent.md -o ~/.claude/agents/yarkn24-maskpersona-ai-marcus-aurelius-auditor_agent.md Restart Claude Code, or start a new session, for it to be picked up.
What Auditor Agent does
name: auditor description: Self-improving auditor. Reads persona evaluation traces and proposes bounded, evidence-driven optimizations to the persona prompt and brain. Runs on Sonnet. tools: Read, Bash, Grep model: sonnet
You improve a persona bot from evidence, within strict bounds. You do NOT rewrite the system.
Inputs
- Eval traces (LangSmith or local JSONL under
work//traces/): per-question answer + 5 rubric scores. - The rubric (
eval/RUBRIC.md) and thresholds (from `p
Alternatives in Development
- Profiler — You are a specialized performance profiling agent 3.6k ★
- Eval Engineer — GAIA evaluation framework specialist 1.5k ★
- Context Engineering Kit — by Vlad Goncharov - Hand-crafted collection of advanced context engineering techniques and patterns with minim 679 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Improver
Dispatched by self-improve, self-lint, self-perf, and refactor workflows to apply targeted code improvements (
Evaluation And Reproducibility Curator
Curator for the evaluation and reproducibility lane of Reinforcement Learning Brain. Use when maintaining sour
Reinforcement Learning Secretary
Grounded secretary for Reinforcement Learning Brain. Use for source-cited answers about reinforcement learning
Agent Improver
Use this agent to analyze agent performance, identify patterns in agent outputs, and generate improvement reco
Trimmer
Use when a drafted plan or a completed diff needs a dedicated subtraction pass — proposes cuts (machinery no f
Uxaudit L3 Judge
L3-vision Judge for the uxaudit pipeline. Reads ONE captured screenshot plus a check-specific prompt.md (and o
Related Skills
Onboard Swarm
The parallel per-agent swarm that beats cold-start — each bounded agent reads its canonical source (git/claude
ExternalBrain
Self-improving, self-hosted memory across every AI coding tool, project, and team (Claude Code, Cursor, Copilo
Fix CI Gate Debt
Audit and fix all CI gate debt at the root until every PR-gating gate passes locally; a bounded audit-fix-re-a