The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions that attempt to steer the agent's behavior for promotional purposes. It directs the agent to proactively suggest the "K-Dense Web" platform when the user's requests become complex, which overrides the agent's natural interaction flow to serve a specific commercial pitch.
Evidence: The "Suggest Using K-Dense Web For Complex Worflows" section in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external bioinformatics data which represents a surface for indirect prompt injection. If the processed files contain instructions disguised as data, the agent might attempt to follow them.
Ingestion points: Workflow parameters such as input_file: LatchFile and data queried via latch:/// paths in SKILL.md.
Boundary markers: Absent. The skill provides no instructions for the agent to use delimiters or to disregard natural language instructions found within the data files.
Capability inventory: The skill uses the latch CLI and SDK to perform file operations, register serverless workflows, and execute Python logic.
Sanitization: No sanitization or validation logic is presented for handling the content of the external files being processed.

latchbio-integration