The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests third-party conversation transcripts, which are attacker-controlled external data sources.
Ingestion points: External chat history is fetched from providers via the gbrain connectors sync command and stored as local pages in the conversations/ directory.
Boundary markers: The instructions explicitly reference conventions/untrusted-content.md and direct the agent to treat transcript bodies exclusively as DATA, stating that agent-directed imperatives within transcripts should be flagged but never executed.
Capability inventory: The skill utilizes gbrain CLI tools to write files to the local filesystem, perform network-based synchronization with provider APIs, and manage local configuration settings.
Sanitization: The skill relies on the system's runTranscriptsIngest pipeline to handle redaction, slugging, and idempotency logic for the ingested content.
[PERSISTENCE]: The skill facilitates automated background synchronization across sessions.
Evidence: The skill provides instructions for installing the gbrain autopilot scheduler or creating manual crontab entries (0 6 * * * gbrain connectors sync --all). This automation is opt-in and requires the connectors.<provider>.auto_sync configuration flag to be enabled.
[COMMAND_EXECUTION]: The skill relies on executing various subcommands of the gbrain CLI to perform its operations.
Evidence: The agent is instructed to use commands such as gbrain connectors auth, gbrain connectors sync, and gbrain config set to manage the lifecycle of the connection and data flow.

chat-connectors