The Agent Skills Directory

[SAFE]: The skill is purely informational, providing a mapping for conventional commit keywords (e.g., 'feat', 'fix') to GitHub labels (e.g., 't:feature', 't:bug').
[NO_CODE]: No executable code, scripts, or remote dependencies are included or referenced for execution. The mention of repository-local files like .github/scripts/label-pr.py is for workflow context only and does not involve runtime execution by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process task descriptions to infer a commit type. This is a low-risk data ingestion surface because the output is restricted to a small set of predefined keywords and labels, with no downstream execution capabilities.
Ingestion points: User-provided task descriptions and change summaries (SKILL.md).
Boundary markers: None defined.
Capability inventory: None detected. The skill does not write files, access the network, or execute commands.
Sanitization: Not required for the simple keyword output provided.

labeling-changes