Menu
Use when creating, revising, validating, or deploying a reusable agent skill.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Consult after receiving review feedback and before implementation when the feedback needs evidence-based evaluation.
Consult before delegating implementation when an approved plan benefits from worker context isolation and durable coordination.
Consult before attempting a fix when root cause or behavior is unclear, failures repeat, components interact, or guessing would be unsafe.
Consult before implementing an observable behavior change when a meaningful automated test seam exists; use proportional red-green evidence.
Consult directly matching task playbooks before substantial design, debugging, implementation, review, or completion work without forcing full workflow execution.
Consult before making a substantive completion claim; require current trustworthy evidence and semantic acceptance.
| name | writing-skills |
| description | Use when creating, revising, validating, or deploying a reusable agent skill. |
Build a skill only when reusable guidance adds value beyond project instructions,
automation, or a one-off note. A skill can be a minimal meaningful artifact:
frontmatter plus a focused SKILL.md, with extra files only when they improve use.
Use a directory named for the skill. Its SKILL.md frontmatter needs a stable
name and a concise description that names concrete triggers, symptoms, and
boundaries. The body explains the decision, semantic examples, and the smallest
safe workflow. Prefer progressive disclosure: keep first-use guidance short and
move heavy reference, scripts, or assets into linked files.
Map instructions to the active platform's tools and deployment surface rather than assuming a namespace or runtime. Validate markdown syntax and links, then resolve how the target harness discovers and loads the skill.
Choose validation proportional to novelty and risk. A reference skill may need static checks and semantic examples; a behavior/process skill may need a real scenario and a counterfactual or no-guidance control when it could change the conclusion. Preserve useful drafts and iterate from evidence rather than deleting work as ritual.
Commit, publish, push, PR, and release behavior follow current repository policy and explicit authority. Do not use announcements, todo compliance, authority, scarcity, or commitment manipulation as substitutes for a clear skill.
