Skills & docs authoring
Create and continuously improve the skills and docs your agents rely on
Every skill and doc you add is reloaded into the agent's context over and over, and verbose ones get skimmed or ignored. These three skills keep your skills and docs compact and effective, and turn every correction into a lasting lesson.
compact-docs-writer
Rewrites a doc, or drafts a new one, to carry every rule and intent in the least text possible. You get a measured word delta proving nothing was lost, and the diff is applied only after your approval.
compact-skill-creator
Creates or edits skills on top of compact-docs-writer, adding the skill-specific craft: a sharp description (the most expensive text, loaded in every session), the right trigger type, and progressive disclosure. Before showing you the draft it checks the skill against itself, and every edit to an existing skill comes with a version bump, so you never have to remember it.
self-improve
When the agent makes a mistake or doesn't behave the way you want, this skill captures the lesson as a diff to the skill or doc that should have prevented it. You approve where it lands, and the mistake doesn't repeat. Other skills feed it too: memory-doctor moves stray memories into the right doc through it, and refine-pr-review, self-review and harden-artifact hand it the lessons from a review.
Opinionated rules that support this approach
Optional and not installed by default: four rules that support this authoring approach.
- compact-governing-docs every edit to a skill or governing doc goes through the compaction skills first. Read the rule →
- read-other-repos-governing-docs governing docs load on their own only in the project you're working in, so before the agent edits another repo, like the one your skills live in, it reads and follows that repo's AGENTS.md or CLAUDE.md. Read the rule →
- self-contained-docs planning docs, and prompts that hand work to a fresh session, carry everything that session needs and nothing more. Prompts are written with /prepare-prompt. Read the rule →
- self-improve-on-correction when you correct the agent on something a doc governs, it offers to capture the lesson with /self-improve. Read the rule →