The Agent Skills Directory

[SAFE]: The skill implements research workflows using specialized tools like Firecrawl and Exa. These operations are consistent with the skill's stated purpose and use standard MCP tool patterns.
[SAFE]: Indirect Prompt Injection Surface Analysis. The skill processes untrusted web content, which presents an inherent surface for indirect prompt injection.
Ingestion points: Data enters the agent context through firecrawl_scrape and crawling_exa tool outputs.
Boundary markers: No explicit delimiters are specified in SKILL.md for the scraped content.
Capability inventory: The skill includes instructions to save research reports to files.
Sanitization: No content sanitization steps are defined for the processed data.
Mitigation: The skill defines 'Quality Rules' requiring citations, cross-referencing, and separation of fact from inference, which effectively mitigates the risks associated with processing external content.
[SAFE]: Configuration paths such as ~/.codex.json are mentioned purely as guidance for the user's local configuration and do not involve automated access or exfiltration by the agent.

deep-research