The Agent Skills Directory

[SAFE]: A comprehensive review of the skill's instructions and reference materials found no evidence of malicious intent or behavior. The workflow is designed with safety in mind, ensuring that the AI agent only acts as a reviewer and planner until the user explicitly authorizes the implementation of changes.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and analyzing external content, which presents a surface for potential indirect instruction injection.
Ingestion points: The skill reads various files including SKILL.md, references/, scripts/, and assets/ from other local skill directories.
Boundary markers: While the skill does not use specific delimiters for data parsing, it treats all ingested data as subject to review.
Capability inventory: The skill possesses the capability to modify or create files on the system during the Implement stage.
Sanitization: The potential for unauthorized action is mitigated by a strict procedural gate; the agent is instructed to wait for explicit confirmation from the user (e.g., "start modifying") before any changes are applied.
[EXTERNAL_DOWNLOADS]: The skill's reference materials contain links to official developer documentation on the platform.claude.com domain. These are well-known, trusted resources used for pedagogical purposes and do not constitute a risk.

skill-optimizer