The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local shell scripts (granola-mcp-client.sh and oauth-setup.sh) using the bash interpreter. These scripts handle core functionality such as searching and retrieving meeting content. Search queries provided by the user are passed as arguments to these scripts, which may pose a risk if the scripts do not adequately sanitize the inputs.\n- [DATA_EXFILTRATION]: The skill accesses local files that store sensitive authentication tokens, specifically ~/.granola-mcp-token and ~/.mcporter/credentials.json. While these files are used to maintain session state with Granola.ai, accessing credential stores from other local tools (like mcporter) increases the risk of sensitive data exposure.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to https://mcp.granola.ai/mcp to interact with the Granola meeting notes platform. This is the primary communication channel for searching and fetching meeting data.\n- [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it retrieves and displays content from external meeting transcripts. Malicious text within a transcript could potentially influence the agent's logic when summarizing or processing the notes.\n
Ingestion points: Meeting transcripts retrieved via get_meeting_content as described in SKILL.md.\n
Boundary markers: Absent. The output contract does not specify delimiters for transcript content or instructions to ignore embedded commands.\n
Capability inventory: Execution of local shell scripts and network access to the Granola API.\n
Sanitization: The skill does not describe any sanitization or filtering of the transcript text before it is returned to the agent.

granola-transcript