The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external meeting recorders (e.g., Granola, Fireflies) and processes user-generated notes in the originals/ folder. This establishes an attack surface where untrusted external content could contain malicious instructions. However, the skill implements defensive conventions such as an explicit --- timeline --- separator to isolate raw evidence from synthesized information and instructs the agent to treat transcripts as verbatim records to minimize unauthorized instruction following.
[COMMAND_EXECUTION]: The documentation mentions an optional external CLI tool (gbrain) for vault maintenance and enrichment. The skill describes how this tool interacts with the folder structure but does not include instructions for the agent to execute arbitrary or dangerous shell commands.

personal-crm