The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it systematically processes external, untrusted content.
Ingestion points: Untrusted data enters the context via mission briefs in skills/receive-mission.md and external web content from the groktocrawl tool and curl in skills/research-gather.md.
Boundary markers: No explicit boundary markers or delimiters are used to wrap external content to prevent the agent from executing instructions embedded in that content.
Capability inventory: The skill uses write_file, shell commands (mkdir, ls, cat), and the groktocrawl autonomous research suite.
Sanitization: Content retrieved from external sources is synthesized without escaping or validation, allowing potentially malicious instructions in research data to influence the agent's behavior.

researcher-workflow