| name | skill-name |
| category | design|safety|reasoning|architecture|workflow|verification|backend|security |
| version | 1.0.0 |
| priority | P1|P2|P3|P4 |
| auto_load | true|false |
| triggers | ["specific_event_or_domain","another_trigger"] |
| description | [REQUIRED — This is what Claude reads to decide whether to apply the skill]
Apply this skill for ANY [exhaustive list of scenarios where this applies].
Trigger verbs: ANY, BEFORE, WHEN, AFTER.
Be specific — vague descriptions cause missed triggers.
End with: "[Priority context]. [Auto-load statement if P1]."
|
| context | fork |
[Skill Display Name]
The Default Being Overridden
[REQUIRED — Name the specific LLM defaults this skill replaces. Be concrete.]
Left unchecked, LLMs default to:
- [Specific bad default #1 — name the exact behavior, not a vague category]
- [Specific bad default #2]
- [Specific bad default #3]
This skill overrides those defaults with [one-sentence summary of replacement approach].
ABSOLUTE RULES (Never Violate)
[REQUIRED — The non-negotiable rules. Use tables, code blocks, or numbered lists.]
[Rules that start with NEVER or ALWAYS are the highest-signal part of this skill.]
NEVER:
ALWAYS:
- [Specific required action]
[Rule Category 1]
[REQUIRED — At least 2 substantive sections with concrete patterns/rules]
[Use code blocks for patterns that should be copied literally]
[Use tables for comparison information]
[Use numbered lists for sequences]
[Use prose for reasoning that requires explanation]
[Rule Category 2]
[Reference materials go here — palettes, patterns, checklist items, command examples]
References
[Optional — Point to related skills, agents, or files]
[e.g.: "See references/color-system.md for full token set."]
[e.g.: "This skill triggers council-of-logic when [condition]."]