Create Agent Skill

Canonical Instructions

Read and follow repository-root AGENTS.md; this skill narrows that contract and never overrides it.

Description And Trigger

Use when adding or revising a reusable skill under .agents/skills/.

Inputs

• concrete repeatable task
• repository constraints
• required artifacts or scripts

Preconditions

• the task is repeatable and repository-driven

Ordered Workflow

1. Define the trigger and inputs.
2. Write an ordered workflow.
3. Specify required outputs, evidence, validation, failure, and escalation.
4. Add only minimal supporting assets, references, or scripts.

Required Outputs

• skill definition
• optional supporting assets or scripts

Evidence

• repeatable procedure
• validation approach

Validation Checklist

• frontmatter contains name and description
• workflow is concrete, not a vague prompt
• failure path is explicit

Failure Conditions

• the skill is only a prompt with no procedure

Escalation Conditions

Escalate if the task is too broad for a reusable skill or depends on unstable hidden context.

Forbidden Behavior

• overloading the skill with unnecessary context
• omitting escalation paths