The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a research pipeline that ingests information from external and internal sources, which inherently introduces a surface for indirect instructions to be processed by the agent.
Ingestion points: Accesses Snowflake metadata, project files, official documentation, internal runbooks, and local document paths as described in the SKILL.md pipeline.
Boundary markers: The process requires a human to approve the retrieval plan before execution and utilizes a Reviewer role that is logically separated from the crawler process.
Capability inventory: No executable scripts or direct subprocess commands are defined within this file.
Sanitization: Not explicitly mentioned in the instructional text.
[SAFE]: The skill incorporates safety-conscious design patterns, specifically the 'Human approves the retrieval plan' step and the separation of concerns where a 'Reviewer' validates citations without exposure to the raw research process. No indicators of data exfiltration, credential theft, or remote code execution were found.

flow-research