Agent Skills: Skill Help

Skill Help

Explain one named local Agent Skill. Inspect only; never execute the target skill.

Workflow

1. Parse parameters from the prompt:
   • -n [NAME] is required. Parse the exact Skill name; when it is absent, return this Skill's usage instructions.
   • -path [PATH] is optional. Parse the supplied path or paths; when absent, set [PATH] to empty. Do not pass -path to the script; it then resolves its default roots in order: SKILL_PATHS, $CODEX_HOME/skills, ~/.codex/skills, ~/.claude/skills, ~/.agents/skills.
   • Infer and directly replace each non-empty [PATH] with its Skill storage directory. Do not call any tool to locate, verify, or further search for that directory. For example, infer and replace C:\Users\abc\tool\skills\skill-help\agents with C:\Users\abc\tool\skills.
2. Run scripts/inspect_skill.py -n NAME --json; pass each parsed path with -path only when [PATH] is not empty.
   • The script returns JSON. On success, it returns status, name, description, path, valid, validation_warnings, sections, headings, searched_sources, warnings, errors, and resources; resources always lists files in scripts, references, and assets. On failure, it returns status, name, searched_sources, results, warnings, and errors; not-found results also include suggestions.
3. Analyze the output from scripts/inspect_skill.py and present the result using the output schema: first answer the user's question directly; then, honestly and rigorously, use only documented evidence to explain the Skill's purpose, its relationship to the author's current task, and the ways it may help. Clearly label inferences, state Not documented for absent information, and do not invent capabilities or task context.

Guardrails

• Read the complete SKILL.md and list direct files under scripts/, references/, and assets/ when helpful.
• Read a directly referenced file only when necessary; reject paths outside the skill root and do not recursively follow references.
• Never run scripts to infer behavior.
• Report documentation/script inconsistencies and require verification before execution.