rules-check
Skill5 starsChanged 52 days agoMITView on GitHub
Find the rules in a project's CLAUDE.md, AGENTS.md, and other instruction files that repeat or contradict the Katharsis style, and suggest an edit or a cut for each. Changes no file without the user's yes. Use when the user asks to check their instruction files against Katharsis, when setup offers it, or when replies seem pulled between Katharsis and a project rule.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- commitSkill
- prSkill
- review-replySkill
- setupSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 28
Read ~/.claude/katharsis/output-styles/katharsis.md and ~/.claude/katharsis/styles/README.md, the installed copies, since those are the …
- Distinctive: line 64
After an approved edit, re-read the file and confirm the text changed as proposed.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Katharsis rules check
The model receives the user's instruction files and the Katharsis style on every turn. A rule that repeats the style spends context for nothing, and a rule that contradicts it leaves the model to choose between them. This skill finds both and proposes an edit for each. It reads freely and writes only what the user approves.
1. List the instruction files
~/.claude/katharsis/scripts/instruction-files.shThe script prints one line per file Claude Code loads for the current folder, as
<kind><TAB><path>: the managed policy, the user's ~/.claude/CLAUDE.md and rules, every
CLAUDE.md, .claude/CLAUDE.md, and CLAUDE.local.md from the root down to this folder, the
project's .claude/rules/, AGENTS.md where Claude Code reads it, CLAUDE.md files in
subfolders, and every file those pull in with @. Read each file it lists. When it prints
nothing, say that no instruction files load here and stop.
2. Read the style the model receives
Read ~/.claude/katharsis/output-styles/katharsis.md and ~/.claude/katharsis/styles/README.md,
the installed copies, since those are the text the model gets. Open a guidance file under
~/.claude/katharsis/styles/ only when a rule names that exchange type's situation, such as a
rule about answering questions or reporting status.
3. Sort every rule
Take each rule in each file and put it in one group:
- Duplicate. It says what the style already says: answer first, stop and ask only for a call that is expensive to undo, no headers on a short reply, a length limit the ceiling already sets. Suggest cutting it, and name the style's line that already covers it.
- Conflict. It asks for something the style forbids or shapes differently: end every reply with a summary, always list next steps, ask before any edit, never use headings. Say what each rule produces in a reply, and suggest which to keep and the edit that removes the conflict.
- Unrelated. Rules about code, tools, the repo, commands, or the domain. Leave them alone and report only their count per file.
A rule that only partly overlaps is a conflict when following both would change the reply, and a duplicate otherwise. Quote each rule exactly as the file has it.
4. Report
Group the report by file, in the script's order, and within a file list the duplicates, then the conflicts. Each item carries the quoted rule, the style's line it meets, and the suggested edit as the exact text to remove or the replacement text. End with the count of unrelated rules per file. A file with no duplicates and no conflicts gets one line saying so.
5. Apply only what the user approves
Change no file in this skill without the user's yes for that edit. Ask per file, and list the
edits for that file in the question so the yes covers exactly those. A file from the user or
managed kind, or any file outside the current project, reaches every project on the machine:
ask for it separately, and say that in the question. A managed file is usually not writable by
the user; report it and suggest they raise the conflict with whoever manages it.
After an approved edit, re-read the file and confirm the text changed as proposed.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

