rc-writing-for-agents

Authors and audits skills, AGENTS.md files, and hooks for agent context efficiency.

19|1|Updated Jun 27, 2026
One-click install
npx skills add https://github.com/rodolfochicone/rc-project --skill rc-writing-for-agents-rodolfochicone
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rc-writing-for-agents
Source: https://github.com/rodolfochicone/rc-project/tree/main/skills/promoted/rc-writing-for-agents
Command: npx skills add https://github.com/rodolfochicone/rc-project --skill rc-writing-for-agents-rodolfochicone

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Agent-facing documents — skills, resident instruction files like AGENTS.md and CLAUDE.md, and hooks — often bloat the context window, duplicate guidance, or fail to trigger reliably. This Skill provides a doctrine and audit procedure for writing and trimming these artifacts so agents follow them predictably. ## Core Features & Use Cases - Skill authoring and auditing: Create new skills from a template, validate name/description metadata with a bundled script, and run a doctrine-plus-spec checklist audit. - Resident file trimming: Apply the rent test and scope ladder to AGENTS.md/CLAUDE.md files so every resident line earns its context cost. - Hook authoring: Write deterministic hooks in the house style with correct event selection, fail-open behavior, and self-tests. - Context budget audits: Inventory skill descriptions, MCP servers, and resident prose against the host's listing budget. - Use Case: A plugin ships 48 skills whose descriptions exceed the context listing budget, so some skills silently stop firing. Use this Skill to audit each description, cut it to the routing essentials, and re-validate until the listing fits. ## Quick Start Ask the agent to audit your SKILL.md or AGENTS.md file against the writing-for-agents doctrine and fix every failing checklist item.

Frequently Asked Questions about rc-writing-for-agents

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

FAQPage Schema
How do I write a good skill description for agent discovery?▼

Front-load a leading word, include one trigger per genuine branch, and keep it under roughly 150-350 characters in third person. Validate the name and description pair with the bundled validate-metadata.py script before shipping.

How do I trim a bloated AGENTS.md or CLAUDE.md file?▼

Run the rent test line by line: each line must change default behavior, apply to most sessions at its scope, and be cheaper resident than derived on demand. Delete or relocate failures down the scope ladder to subtree files, skills, or hooks.

Why does the agent ignore my skill's references folder?▼

Inline summaries that duplicate reference content let the agent skip the files. Trim inline content to a short tripwire, give each reference exactly one pointer worded with its trigger condition, and escalate must-fire pointers to read-in-full directives.

When should a rule become a hook instead of prose?▼

A rule that must hold every time and can be detected mechanically belongs in a hook, since prose is advisory and gets skipped under context pressure. Rules needing judgment about intent stay in skills or resident files.

What is the skill description listing budget in Claude Code?▼

Claude Code caps skill metadata at about 1% of the context window, roughly 2k tokens on a 200k window. Past the cap it drops descriptions starting with the least-invoked skills, leaving them present by name but unable to fire.