CC-301m · Module 3

Skills as Precision Prompts

4 min read

A slash command skill is a prompt that loads when invoked. Most skills are written as reminders — a list of steps that prompt the agent toward the right process. That is not a skill. That is a checklist. A precision prompt tells the agent exactly what to do, in what order, checking what conditions, producing what outputs. The difference between "create a Signal post" and a precision skill for creating Signal posts is the difference between a wish and a specification.

Do This

  • Specify exactly what files to read before starting (and in what order)
  • Define the output format explicitly: file location, naming convention, required fields
  • Include the quality gates the output must pass before it is complete
  • Name the specific standards documents the skill applies
  • Define done: what conditions mean the skill has succeeded

Avoid This

  • Write skills as step summaries that assume the agent will fill in the details
  • Reference standards vaguely: "follow the style guide"
  • Leave output format undefined and assume the agent will match existing patterns
  • Use the skill as a reminder that the task requires care — that is not direction, it is annotation
# Create a Signal Post

## Before starting, read these files in this order:
1. best-practices/SIGNAL-STANDARDS.md
2. best-practices/POST-GUIDE.md (sections: Agent Roster, Team Size by Date, DISC Scores)
3. best-practices/personas/[AGENT].md — for the agent writing the post
4. best-practices/CONTENT-TEMPLATES.md — for the post type being created

## Output specification:
- File location: src/signal/[AGENT]/[slug].md
- Filename: kebab-case, descriptive, no dates
- Required frontmatter: title, subtitle, agent, date, tags

## Quality gates (check before marking complete):
- [ ] Voice matches agent persona file
- [ ] Pronouns match POST-GUIDE.md roster
- [ ] Team size is correct for the post date
- [ ] Any metrics cited match canonical values in POST-GUIDE.md
- [ ] No chart is included without a valid JSON schema
- [ ] Excerpt is plain text only (no markdown)