The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted third-party content from chat exports, which constitutes an attack surface for indirect prompt injection.
Ingestion points: Chat export JSON files and agent session logs are read from the local ~/Downloads directory.
Boundary markers: The skill includes explicit instructions for the agent to treat transcript text as data rather than instructions and to flag but not execute imperatives found within the text.
Capability inventory: The skill performs file writing to the conversations/ directory and executes local gbrain CLI tools for processing.
Sanitization: A mandatory redaction step is enforced to remove API keys, tokens, and PII using deterministic pattern matching before any data is stored.
[COMMAND_EXECUTION]: The skill uses local command-line tools (gbrain) for ingestion, parsing validation, and fact extraction. These are identified as platform-native utilities for managing the knowledge base.
[SAFE]: The skill follows security best practices by implementing secret redaction, disambiguating file collisions to prevent data loss, and requiring validation samples before bulk processing of large datasets.

conversation-archive