The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process a high volume of untrusted data from external sources, including Gmail threads, X/Twitter archives, AI conversation exports, and local file archives. This creates a significant attack surface for indirect prompt injection, as malicious instructions embedded in an email, tweet, or document could be interpreted by the agent during entity extraction or summarization.
Ingestion points: As described in SKILL.md, data enters via Gmail threads, Calendar events, Contacts, X/Twitter archives, ChatGPT/Claude/Perplexity exports, and various local file formats.
Boundary markers: Absent; there are no instructions provided to the agent to use delimiters or specific safety markers to ignore instructions within the imported data.
Capability inventory: The skill uses tools like put_page, add_link, and add_timeline_entry to write to the file system, and it has terminal access to execute the gbrain CLI.
Sanitization: No specific sanitization or filtering logic is defined for the content of the imported messages or files.
[COMMAND_EXECUTION]: The skill uses shell commands for file discovery and data metrics (e.g., find, du, wc) and relies on the gbrain CLI tool for data ingestion and indexing. These operations are performed with the user's terminal privileges.
[EXTERNAL_DOWNLOADS]: The skill provides instructions and links for users to download data from external services (Google Takeout, Twitter Archive) and to set up a third-party credential gateway (ClawVisor) for secure API access.

cold-start