The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted meeting transcripts and uses the extracted data to perform file operations, creating a surface for indirect prompt injection attacks where malicious instructions in a transcript could influence agent behavior.
Ingestion points: Meeting transcripts are read from the file system in scripts/process.py and passed to the LLM extractors.
Boundary markers: The LLM prompts in scripts/extractors/leadgen.py, scripts/extractors/partnership.py, and scripts/extractors/coaching.py use simple text headers (e.g., "Transcript:") to delimit untrusted content, which is insufficient to prevent adversarial instructions from being interpreted as commands by the model.
Capability inventory: The skill can make network calls to api.cerebras.ai, write to local files (appending analysis), and perform file system searches and updates (linking prep notes as described in SKILL.md).
Sanitization: There is no evidence of sanitization, filtering, or escaping of the transcript content before it is interpolated into prompts.
[COMMAND_EXECUTION]: The skill involves executing shell commands for setting up the environment and performing file system tasks as part of its normal workflow.
Evidence: SKILL.md and README.md include instructions for the agent to run pip install for dependencies and find commands to locate and link related meeting notes.

meeting-processor