The Agent Skills Directory

[SAFE]: The skill establishes a high-trust research framework with specific guidelines for sourcing and evidence validation. It prioritizes authoritative records and direct observations while maintaining a clear research contract for provenance.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges a significant data ingestion surface (analyzing external technical research and upstream source code) but incorporates robust defensive guidelines. It explicitly directs the agent to treat all upstream content as untrusted evidence and ensures that instructions found within inspected projects do not override the current task's authority. 1. Ingestion points: External upstream repositories, package manager data, and web research sources (references/exact-source-grounding.md). 2. Boundary markers: The skill relies on clear research pinning rather than specific character delimiters. 3. Capability inventory: Access to browser-rendered sources, repository providers, package managers, and local artifacts. 4. Sanitization: Mandatory guideline to treat all ingested code as untrusted and maintain task-level authority (references/exact-source-grounding.md).

iwadi