The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to explore and read source code across multiple packages and directories to map the existing landscape. This creates a surface for indirect prompt injection, where an attacker could place malicious instructions within code comments or metadata in the files being analyzed.\n- [INDIRECT_PROMPT_INJECTION]: Ingestion points: Phase 1 (Deep Exploration) requires the agent to read key files across the target package, reference implementations, and consumer codebases.\n- [INDIRECT_PROMPT_INJECTION]: Boundary markers: The instructions lack explicit delimiters or mandatory warnings to ignore embedded instructions found within the source code being analyzed.\n- [INDIRECT_PROMPT_INJECTION]: Capability inventory: The skill utilizes file reading, the AskUserQuestion tool, and markdown file writing across the project directory.\n- [INDIRECT_PROMPT_INJECTION]: Sanitization: There is no requirement or guidance for the agent to sanitize, escape, or filter external content before incorporating findings into the final specification.

creating-spec