The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection by ingesting and acting upon untrusted data. 1. Ingestion points: The skill reads external project states, including files, documentation, plans, and prior decisions (SKILL.md). 2. Boundary markers: The process uses a mandatory 'Understanding Lock' phase and incremental 'Does this look right?' checkpoints as logical boundaries to confirm user intent. 3. Capability inventory: The skill writes finalized design specifications and decision logs to durable Markdown files. 4. Sanitization: The instructions do not mention explicit sanitization, filtering, or escaping of the content ingested from the project files.

brainstorming