The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from untrusted external sources, which is a common vector for indirect prompt injection attacks where malicious instructions are embedded in web content to influence the agent.
Ingestion points: Data retrieved from WebSearch, WebFetch, and the Playwright browser automation tool, as described in SKILL.md.
Boundary markers: The methodology explicitly instructs the agent in SKILL.md to "Cross-reference everything," "Verify facts appear consistently," and "Be skeptical of content farms" to mitigate the risk of relying on a single malicious source.
Capability inventory: The agent uses network-based tools for research and local tools (Grep, Glob, Read) for codebase exploration as identified in the Available Research Tools section.
Sanitization: The skill relies on logical verification, manual source quality assessment, and the CRAAP Test methodology defined in references/source-evaluation.md rather than programmatic data filtering.

deep-research