name: <kebab-case-name> description: <one-sentence trigger condition — when should this skill be loaded? Start with "Apply when ..." for consistency.> license: MIT version: 1.0.0 tokens_target: 1500 triggers:

• <short trigger phrase>
• <another phrase the router matches against>

loads_after: [] supersedes: []

Sub-Skill: [Name]

<!-- target: tokens_target above. Run python tools/render_readme_table.py to update README. -->

Purpose: [1-2 sentences. What this sub-skill does and why it exists.]

Rules

[Category 1]

1. [Rule name]. [Action directive — must start with Always, Never, Before, After, Prefer, Avoid, Use, Do, or Ensure.]
2. [Rule name]. [Action directive.] Reference: ERR-YYYY-NNN
3. [Rule name]. [Action directive.]

[Category 2]

1. [Rule name]. [Action directive.]
2. [Rule name]. [Action directive.]

See also

• skills/code-quality/SKILL.md (when this skill specialises a generic rule, link to it)
• skills/error-log/SKILL.md (errors that motivated rules in this skill)

Notes

• Frontmatter is mandatory. The CI validator (tools/validate_rules.py) will fail any sub-skill without a valid frontmatter against schemas/skill-manifest.json.
• Action directives, not descriptions. "SQL injection is dangerous" is a description; "Never concatenate user input directly into SQL queries; always use prepared statements" is an action directive.
• `Reference: ERR-YYYY-NNN` on every rule born out of a real observed mistake.
• Token budget under 3000. When exceeded, split into thematic sub-skills with loads_after chains.