The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a sequence of shell-style commands using the gbrain CLI (think, search, get, query, timeline, backlinks, graph) to access and synthesize internal information. These commands use templates that interpolate entity names, which could represent a command injection risk if the entity strings are not properly sanitized before being passed to a shell environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, including email archives, meeting transcripts, and web search results, to identify relationships and update internal records.
Ingestion points: The skill reads data from mounted sources (Step 3) such as email archives and chat transcripts, as well as external web search results (Step 5).
Boundary markers: There are no explicit boundary markers or instructions defined to prevent the agent from following malicious instructions that might be embedded in the ingested emails or web pages.
Capability inventory: The skill has the capability to read sensitive internal data, perform network operations (web search), and write modifications to the agent's persistent knowledge base in the people/ and companies/ directories.
Sanitization: There is no evidence of sanitization or filtering of the external content before it is processed by the think synthesis tool or written to internal pages.

resolve-before-asking