The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process AI-facing instruction files (such as CLAUDE.md, SKILL.md, and agent definitions) for optimization. This creates an attack surface where instructions embedded in the target files could influence the agent's behavior during the audit process.
Ingestion points: The "Optimization Checklist" in SKILL.md requires the agent to read and audit external target files (specifically in Pass 1, Pass 3, and Pass 4).
Boundary markers: The instructions do not provide specific delimiters or isolation protocols to distinguish the content being audited from the agent's own control instructions.
Capability inventory: The skill allows the agent to read files, create file backups, and dispatch subagents to gather information and validate instructions.
Sanitization: No explicit sanitization or filtering of the ingested instruction content is described before it is processed by the agent or its subagents.
[EXTERNAL_DOWNLOADS]: The reference files (references/agents.md, references/memory-and-rules.md, and references/skills.md) cite source materials and guidelines hosted on Anthropic's official GitHub repositories. These are used to provide a baseline for optimization best practices.
[COMMAND_EXECUTION]: The skill instructs the agent to utilize CLI tools and discovery commands (e.g., --help, gh, and project-specific scripts) to verify discoverable data at runtime instead of relying on hardcoded facts in instruction files.

optimize