The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes meeting notes and transcripts retrieved from Granola's servers. These external data sources could potentially contain embedded instructions intended to influence the agent's behavior.
Ingestion points: Meeting content is ingested via query_granola_meetings and get_meeting_transcript as described in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters (e.g., XML tags or triple quotes) to separate untrusted meeting content from the agent's system instructions.
Capability inventory: The skill allows the execution of granola-cli commands, including the ability to call various MCP tools.
Sanitization: No explicit sanitization or filtering of the meeting content is defined in the instructions.
[COMMAND_EXECUTION]: The skill interacts with the system using the granola-cli binary. While it executes shell commands, these are restricted to the subcommands defined for managing Granola connections and tool calls.
[SAFE]: The authentication process uses OAuth with PKCE, and instructions explicitly forbid the agent from accessing or editing sensitive credential files, following security best practices.

granola