The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes vast amounts of untrusted data from multiple external sources, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Importing Gmail mbox files, Twitter archives, AI conversation exports (ChatGPT/Claude), meeting transcripts, and external Markdown files.
Boundary markers: The instructions lack specific guidance on using delimiters or "ignore instructions" wrappers when processing and storing external content.
Capability inventory: The skill has extensive capabilities, including executing shell commands (find, du, wc), using CLI tools (gbrain), and writing to the local filesystem (put_page, sync_brain).
Sanitization: There is no mention of sanitizing or filtering instructions embedded within the imported content, which is stored as "original thinking" without paraphrasing.
[COMMAND_EXECUTION]: The skill uses shell commands and scripts for discovery and processing, such as scanning directories for markdown files and calculating data sizes.
Evidence: Execution of find, du, wc, and gbrain CLI commands in Phase 1 and Phase 0 scripts.
[DATA_EXFILTRATION]: While the skill's primary purpose is local data ingestion, it interacts with an external service (clawvisor.com) for credential management and encourages setting environment variables with API tokens. The skill also facilitates the movement of sensitive personal data (emails, calendar, contacts) from cloud services to a local repository.

cold-start