void-claude-md

Write and audit lean, runnable CLAUDE.md files for coding agents.

Updated May 29, 2026
One-click install
npx skills add https://github.com/voidcorp-core/void-harness --skill void-claude-md-voidcorp-core
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: void-claude-md
Source: https://github.com/voidcorp-core/void-harness/tree/main/packages/core/skills/void-claude-md
Command: npx skills add https://github.com/voidcorp-core/void-harness --skill void-claude-md-voidcorp-core

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? CLAUDE.md files are injected into the agent's context with a "may or may not be relevant" framing, so bloated files dilute attention and bury the rules that matter. This Skill provides doctrine for authoring and pruning CLAUDE.md (or AGENTS.md) files so every line earns its slot. ## Core Features & Use Cases - Authoring rules: Defines what belongs in CLAUDE.md (universal commands, architecture orientation, unenforced invariants) and routes everything else to linters, hooks, settings.json, or deferred docs. - Audit checklist: Provides a pruning checklist built around the question "would removing this line cause an error?" to shrink existing files. - Progressive disclosure patterns: Shows how to point to deeper docs and scope conditional guidance with narrow <important if="..."> blocks. - Use Case: After running /init produces a 400-line CLAUDE.md, use this Skill to prune it to ~60 runnable lines, moving style rules to a formatter and certainties to hooks. ## Quick Start Ask the agent to audit the project's CLAUDE.md and prune every line that would not cause an error if removed.

Frequently Asked Questions about void-claude-md

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a good CLAUDE.md file?

Keep it near 60 lines with only universal, load-bearing content: build/test/dev commands, a one-sentence architecture orientation, and invariants with no mechanical enforcer. Defer detail to referenced docs and verify every command runs from a clean checkout.

What should not go in a CLAUDE.md file?

Exclude code style and formatting rules (use a linter), code snippets (use file:line references), every-time certainties (use hooks), deterministic config like permissions (use settings.json), and deep process docs (defer to referenced files).

How do I audit or prune an existing CLAUDE.md?

Walk every line asking "would removing this cause an error?" and cut anything that fails. Move style rules to linters, certainties to hooks, snippets to file:line references, and occasional detail to deferred docs with a pointer.

Why is my CLAUDE.md being ignored by the agent?

CLAUDE.md is injected as context the model weighs, not a contract it must obey, so long files dilute attention and bury key rules. The fix is fewer, sharper, runnable lines plus mechanical enforcement through hooks and settings.

When should a rule be a hook instead of a CLAUDE.md instruction?

Any rule that must hold 100% of the time belongs in a hook, because instructions are obeyed probabilistically while hooks fire deterministically on every tool call. CLAUDE.md should only carry invariants that have no mechanical enforcer.