skills.md — Template

Summary

Reusable how-to procedures the agent can invoke. Each skill is a named, step-by-step recipe that has been proven to work (promoted from the decision log). Always iterated. See skills.md.

# Skills — <agent name>
version: 0.1
last_updated: 2026-06-05
 
## Skill: produce_slide_carousel
when_to_use: turning a content brief into a TikTok slide carousel
inputs: content_brief, brand_assets
steps:
  1. Read brief + load policies.md limits
  2. Draft hook slide (style per policies.hook_style)
  3. 1 idea per slide, slide_count within policies range
  4. Add source citation slide + disclaimer
  5. Apply brand design system
  6. Save to /content/<account>/<date>/ and update tracker
output: carousel files + tracker row (status: produced)
proven_by: [[Decisions 2026-06#-|... ]]   # link the evidence
 
## Skill: write_hook
when_to_use: first line / first slide
steps: ...
proven_by: ...
 
## Skill: format_for_RAG
when_to_use: adding a source to the knowledge base
steps:
  1. One atomic note per concept
  2. Frontmatter: type, tags, source, date
  3. Clear headings, short chunks
  4. Tag evergreen vs transient

Promotion rule

A procedure only earns a place in skills.md once a decision-log entry shows it worked. Link the evidence with proven_by. This keeps skills.md trustworthy instead of aspirational.

Related: policies.template · guardrail.template · Part A1 - Agent Operating Files