Whyrule — Development skill for Claude Code
Claude isn't ignoring your CLAUDE.md — it never got it.
How to install Whyrule
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open hjalti-hub/whyrule and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Whyrule does
Claude isn't ignoring your CLAUDE.md — it never got it. Finds instructions that never load, imports that resolve to nothing, text stripped before injection, and rules that silently contradict each other. No API key, no dependencies, and it installs itself as a hook.
Alternatives in Development
- Claude Instructions For Local Skill Loading — This file tells Claude how to load skills locally 845 ★
- Pi Reserved Keybindings — Extensions cannot override these shortcuts — they are silently skipped by registerShortcut() 526 ★
- Bm Doctor — Diagnose and recover basemind when it isn't working (MCP tools missing/erroring, "no index", dead server) — ru 94 ★
README
whyrule
Claude isn't ignoring your CLAUDE.md. It never got it.
You wrote the rule down. You wrote it clearly. Claude does the opposite anyway, and there's no error, no warning, nothing in the logs — so you rewrite the rule in capitals and hope.
Usually the rule never arrived. Things that make text vanish between your file and the model, none of which are visible by reading the file:
- an instruction inside an
, which is stripped before injection @docs/setup.mdpointing at a file that moved — imports that resolve to nothing load nothing, and say nothing- an
AGENTS.mddoing the work, which Claude Code does not read - a
.claude/rules/file whosepaths:globs match no file in the repo, so it never loads - your project rule and your
~/.claude/CLAUDE.mdrule flatly contradicting each other — they're concatenated, not overridden, so Claude picks one arbitrarily
`whyrule` finds all of it. There is nothing to install to try it:
git clone https://github.com/hjalti-hub/whyrule && cd whyrule
python3 -m whyrule
No API key. No model calls. No dependencies. Just `python3`.
Once you want it to run without being asked, install it properly and it [hooks itself in](#running-itself) — after that you never type it again.
$ whyrule
CLAUDE.md
18: error REACH001 Instruction inside an HTML comment never reaches Claude: '- Always use `pnpm`, never `npm`, since the lockfile is p...'
fix: Move the instruction outside the comment. Keep comments for notes meant only
for human maintainers, which is what the stripping is for.
3: error IMPORT001 `@docs/architecture.md` imports docs/architecture.md, which does not exist - nothing is loaded
fix: Point it at a file that exists. Relative paths resolve against ./, the
directory of the file the import is written in - not the working directory.
4: error IMPORT001 `@CONVENTIONS.md` imports CONVENTIONS.md, which does not exist - nothin
Related Skills
Axiomarium
See exactly which instructions your coding agent loads for any file, and what it silently drops. axm resolves
Cua Hunt
Hunt Computer-Use-Agent (CUA) prompt-injection surface — instructions planted in CUA-readable DOM channels (hi
Bifrost Debug
Diagnose why the Bifrost gateway, memory injection, or skill discovery isn't working.
Squad Resume
Resume the most recent in-progress happysquad work — scans all state files, finds the latest run/brainstorm/fl
Plan Doctor
Self-check for the planning-with-files mechanisms that fail silently: plan resolution, hook injection, canonic
Skills Janitor
Tinder for your Claude Code skills, subagents and MCP servers. Swipe away what wastes context, scan for prompt
Related Agents
Execution
Reviews whether the task actually got done — plan followed, work finished rather than finished-looking, nothin
Baseline Reviewer
Read-only reviewer for changes to this plugin — skills, agent definitions, manifests, hooks and eval cases. Ch
Triage Analyst
Root-cause analyst. Use when something is broken, erroring, hanging, silently doing nothing, or returning wron