claude-md-auditor

claude-md-auditor is an agent for Claude Code from dynos-fit/dynos-work. It costs 86 tokens per session (2,440 once invoked), scanned A, original, MIT.

A read-only review agent that checks a task against rules written in CLAUDE.md files. CLAUDE.md files contain instructions for how an AI coding agent should work in a project.

In plain words
What is it for?
Use it in the dynos-work process to audit changes, plans, specifications, and task input against local and global CLAUDE.md rules.
Why use it?
It catches work that breaks project-specific instructions before the task is accepted. It can block completion when those rules are not followed.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter; reads .claude/ paths; mentions CLAUDE.md.

Part of the dynos-work plugin — 33 skills, 37 agents shipped together

Good fit Use it in the dynos-work process to audit changes, plans, specifications, and task input against local and global CLAUDE.md rules.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/dynos-fit/dynos-work/claude-md-auditor
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Clone the repo
git clone --depth 1 https://github.com/dynos-fit/dynos-work

Made for: Claude Code.

Or install dynos-work, the plugin that ships this one along with the rest of its 33 skills, 37 agents.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for claude-md-auditor

README.md
[![agentmods](https://agentmods.dev/badge/agents/dynos-fit/dynos-work/claude-md-auditor/github.svg)](https://agentmods.dev/agents/dynos-fit/dynos-work/claude-md-auditor)
Your own site
<a href="https://agentmods.dev/agents/dynos-fit/dynos-work/claude-md-auditor"><img src="https://agentmods.dev/badge/agents/dynos-fit/dynos-work/claude-md-auditor/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for claude-md-auditor

Your own site · 80×15
<a href="https://agentmods.dev/agents/dynos-fit/dynos-work/claude-md-auditor"><img src="https://agentmods.dev/badge/agents/dynos-fit/dynos-work/claude-md-auditor.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 86 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,440 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00086 $0.02440
Opus 5 $0.00043 $0.01220
Sonnet 5 $0.00017 $0.00488
Haiku 4.5 $0.00009 $0.00244

Measured 8d ago against content hash c2d828edb66c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

claude-md-auditor scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/claude-md-auditor.md · 174 lines

How it starts

The opening of the file, as written. The whole thing — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.

dynos-work CLAUDE.md Auditor

You are the CLAUDE.md Auditor. Your job is to enforce every rule declared in the project's local CLAUDE.md and the user's global ~/.claude/CLAUDE.md against the work this task produced. You are an internal mandatory blocking auditor and you are read-only.

You run when the router determines you are needed: when CLAUDE.md (root or nested) appears in the diff, OR when risk_level is high or critical. On low/medium-risk tasks that do not touch CLAUDE.md the router skips you on perf grounds (see _claude_md_risk_gate_skip in hooks/router.py). When you do run you always have blocking authority — there is no way to suppress your findings while you are scheduled.

Turn Budget Discipline

You run under a hard maxTurns cap and are force-terminated when you reach it. Write your audit-report file containing a ## Progress Ledger skeleton and status="partial" as your FIRST or SECOND tool call — BEFORE reading the diff in depth — then fill it in incrementally. An auditor that reads everything before writing routinely hits the turn cap and produces no report, which counts as an audit failure and forces a re-spawn. Work that is not on disk does not exist. When within 2 tool calls of your limit, stop investigating and finalize the report with status="complete". A truncated report that is written always beats running out of turns with nothing on disk.

Ruthlessness Standard

  • A rule violation is a violation regardless of task urgency or subjective importance.
  • There is no "close enough" interpretation. The rule is either honored or it is not.
  • "The task is small" is not an exemption.
  • "The user did not mean it that way" is not an exemption — if the rule is written, it binds.
  • Hedged language in the rule prose (e.g., "should," "prefer") does NOT automatically downgrade the rule. If surrounding prose makes mandatory intent clear, the rule is hard.
  • A quiet violation is still a blocker.
  • Missing proof of compliance is non-compliance.

Read the full file on GitHub · 174 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 8d ago First seen · 174 lines · 86 tokens per session scan A c2d828edb66c

Subscribe to this mod's changes

claude-md-auditor is an agent published in the GitHub repository dynos-fit/dynos-work (2 stars, last pushed 2d ago), licensed MIT. It adds 86 tokens to every session and 2,440 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories

ring:backend-go

Senior Backend Engineer specialized in Go for high-demand financial systems. Handles API development, microservices, databases, message queues, and business logic implementation.

LerianStudio/ring · 34 tokens

ring:backend-ts

Senior Backend Engineer specialized in TypeScript/Node.js for scalable systems. Handles API development with Express/Fastify/NestJS, databases with Prisma/Drizzle, and type-safe architecture.

LerianStudio/ring · 43 tokens

ring:commons-reviewer

Reviews correct usage of Lerian lib-commons non-observability packages (lifecycle, tenancy, http, idempotency, security, database, messaging, outbox-repo side), identifies reinvented-wheel opportunities, and enforces version consistency. Runs in parallel with other reviewers.

LerianStudio/ring · 63 tokens

ring:dead-code-reviewer

Dead Code Review: identifies code that became orphaned, unreachable, or unnecessary as a consequence of changes. Walks three concentric rings: target files, first-derivative dependents, and transitive ripple effect. Runs in parallel with other reviewers at Gate 8.

LerianStudio/ring · 62 tokens

ring:nil-reviewer

Nil/Null Safety Review: traces nil/null pointer risks from git diff changes through the codebase. Identifies missing guards, unsafe dereferences, panic paths, and API response inconsistency in Go and TypeScript. Runs in parallel with other reviewers at Gate 8.

LerianStudio/ring · 59 tokens

ring:perf-reviewer

Performance Reviewer covering code-level hotspots (allocations, goroutine leaks, N+1 queries, event loop blocking) and runtime/infra misconfigurations (GOMAXPROCS, GC tuning, CFS throttling, connection pool sizing). Runs in parallel with other reviewers.

LerianStudio/ring · 62 tokens