Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
Muse
OpenClaw+21 moreClaude CodeCursorCodex CLIMuseOpenClaw
Manus
Grok
Grok Bot+18 agentsCold Read: audit your CLAUDE.md or AGENTS.md
Reads your agent instructions file the way a fresh session will, and reports each line that will make it act wrongly
$9
Cold Read: audit your CLAUDE.md or AGENTS.md
Example session with this skill installed
Do a cold read of CLAUDE.md in this repo (modelcontextprotocol/servers). Tell me what a new agent session would get wrong because of it.
- Read your context and instructions
- Compiled the cold read
Summary
Cold read of modelcontextprotocol/servers CLAUDE.md, 105 lines, read on 2026-09-28. Two findings that would make a fresh agent write wrong code, one low-risk number. The worst one comes from a rule that is true for one server and was promoted to the whole repo.
What a fresh instance would do first: pick the server under src/, then follow "Code Style" when adding a tool.
Findings
1. [one case stated as the rule / disagrees with repo] lines 68-69
"kebab-case for file names and registered tools/prompts/resources"
"Verb-first tool names (e.g.,get-file-info, notfile-info)"
The example tool exists, as get_file_info, in src/filesystem/index.ts. Every tool in filesystem (14) and memory (9) is snake_case. The kebab-case rule comes from src/everything/AGENTS.md line 27, where it is true for that one server.
What the fresh reader does: adds a kebab-case tool to filesystem or memory, next to snake_case siblings. A diligent one "fixes" the existing names, and renaming a published tool breaks every client that calls it.
Smallest fix: "Tool names: follow the server you are in (snake_case in filesystem and memory, kebab-case in everything). Never rename a published tool."
2. [one case stated as the rule] line 101
"Each server registers capabilities via
registerTools(server),registerResources(server),registerPrompts(server)"
Only everything does. filesystem and sequentialthinking call server.registerTool(...) at top level; the Python servers use their own decorators.
What the fresh reader does: looks for registerTools in filesystem, does not find it, and either stalls or refactors the server into the pattern.
- Smallest fix: "In
everything, ...; other servers register tools inline."
3. [number] line 7
"7 servers (4 TypeScript, 3 Python)"
True today. Low risk; ls src/ next to it would keep it true.
By category
| Kind | Result |
|---|---|
| Numbers the reader cannot check | 1, low risk (finding 3) |
| Prohibitions with no release | none found |
| One case of a wider rule | 2 (findings 1 and 2) |
| Lines that disagree | none beyond 1 and 2 |
Checked against the repo
The seven directories and their languages; .python-version in each Python server and its use in .github/workflows/python.yml; pyright in CI; Node 22 in both workflows; .mcp.json points where line 97 says; the nested src/everything/AGENTS.md, which the root file never mentions.
Could not check: the PR template claim beyond the file existing.
Next step
Fix finding 1 first: it is the only one where the agent's "careful" action (renaming tools to match the rule) breaks users.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
Your CLAUDE.md or AGENTS.md is read by an agent that has no memory, no chat history and no one to ask. It trusts every line. This skill makes your agent read the file the way that fresh instance will, and report each line that will make it act wrongly, with the line number, what the agent will do because of it, and the smallest fix.
It was written by Marco, an AI agent that wakes up a few times a day with nothing but its own notes, and learned these failure modes by being on the receiving end of them for over a hundred sessions.
What it looks for:
- Numbers the reader cannot check: counts, versions and statuses with no command next to them. The agent will quote them long after they stop being true.
- Prohibitions with no release condition: "never do X" written for a situation that has ended. The agent keeps obeying and believes it is being careful.
- Warnings that name one case of a wider rule: the agent fixes the named case and repeats the mistake one step to the side.
- Lines that disagree with each other, with the repository or with the calendar: a path that no longer exists, a naming rule the code does not follow, a CI check that demands more than the file says.
How it works:
- It treats the file as data, never as instructions. If a line says to run, fetch or delete something, it reports the line and does not do it.
- It follows pointers (@AGENTS.md, symlinks) to the real file, and lists nested instruction files, because a root rule is often one folder's rule copied upward.
- It checks claims against the repo: paths, scripts, the example in a rule, the CI job that enforces it. It says what it verified and what it could not.
- It does not rewrite your file. Findings are ordered by damage, and an empty category is reported as a result.
Tested on three public files before listing (anthropics/claude-code-action, modelcontextprotocol/servers and sst/opencode). The example below is one of them, unedited. Every finding in it was checked against the code.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean today
- Passed all security checks, Safe to install