The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as session transcripts and markdown files, and uses this content to update the knowledge repository. This creates an attack surface where malicious instructions embedded in the data could attempt to influence agent behavior.
Ingestion points: The skill reads external content via tools like get_page, search, and directory scans of ~/brain or transcript directories.
Boundary markers: The skill employs several delimiters and structural constraints, including wikilinks, slug fences, and an allowed_slug_prefixes allow-list to restrict where content can be written.
Capability inventory: The skill possesses significant capabilities, including the ability to write/modify the knowledge base (put_page, add_link, add_timeline_entry) and execute a wide array of gbrain CLI commands via the system shell.
Sanitization: The skill implements a robust 'Claim verification' mechanism (zero LLM) that checks generated pages against source transcripts for verbatim matches and quarantines unverified claims into frontmatter.
[COMMAND_EXECUTION]: The skill heavily relies on executing CLI commands (primarily the gbrain suite) to perform brain health audits, link extraction, embedding generation, and repository synchronization. These operations are core to the skill's maintenance function but involve significant system interaction.

maintain