The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/granola.py uses the macOS security utility to access the Keychain. This is required to retrieve the encryption key (DEK) used to decrypt the Granola app's local session files and extract the API access token.
[COMMAND_EXECUTION]: Various scripts utilize subprocess.run to call system and CLI tools:
curl is used for authenticated interactions with the Granola API.
gdoc is used to search for and edit Google Documents.
cmark-gfm is used to convert Markdown summaries into styled HTML.
git and rsync are used within the update scripts.
[EXTERNAL_DOWNLOADS]: The skill includes self-update scripts (scripts/check-update.sh and scripts/update-skill.sh) that fetch version metadata and source code from the author's public GitHub repository (HartreeWorks/skill--summarise-granola). This is a transparent mechanism to keep the skill's logic up to date.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted meeting transcripts.
Ingestion points: Raw transcript text from data/transcripts/ is interpolated into agent prompts for the summarization (Step 3) and tidying (Step 8) workflows.
Boundary markers: The prompts currently lack explicit boundary markers or instructions to the LLM to disregard any directives found within the transcript text.
Capability inventory: The skill environment includes tools for network communication (curl, gdoc), file system modification, and credential retrieval (security).
Sanitization: There is no evidence of sanitization or escaping of the transcript content before it is processed by the agents.

summarise-granola