The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from external URLs, creating a surface for indirect prompt injection where malicious instructions could be embedded in the ingested content.
Ingestion points: Phase 1 fetches content from user-provided links, articles, and tweets via web fetch tools.
Boundary markers: The instructions do not specify the use of delimiters (e.g., XML tags or triple quotes) or explicit instructions to the agent to ignore embedded commands within the fetched content.
Capability inventory: The skill utilizes tools for database modification (put_page, add_link) and executes local shell commands (gbrain files upload-raw, gbrain sync).
Sanitization: There is no evidence of content sanitization, escaping, or validation before the fetched data is analyzed or passed to the CLI tools.
[COMMAND_EXECUTION]: The skill instructs the agent to perform operations using a local CLI utility named gbrain. These operations include gbrain files upload-raw for data provenance and gbrain sync for indexing. These commands interact with the local file system and environment to manage the knowledge base.

idea-ingest