The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from multiple directories (people/, meetings/, daily/, etc.) where external or untrusted data might reside. Since the agent is instructed to read these files completely and make judgment calls on sanitization, malicious instructions hidden within those files could potentially influence the agent's behavior during the sanitization process.
Ingestion points: Files in people/, meetings/, daily/, companies/, projects/, and analysis/ are read for processing.
Boundary markers: None specified for the content being processed to distinguish data from instructions.
Capability inventory: The skill uses gbrain CLI tools, git (including history rewriting and force-pushing), grep, rsync, and basic filesystem commands (rm, mkdir, cp, find).
Sanitization: Relies on model judgment and keyword triaging; no explicit input sanitization for the processed data is mentioned.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands for file discovery, triaging, and repository management. This includes high-risk operations like git filter-repo and git push --force. While these are gated by confirmation prompts and intended for the skill's primary purpose of sanitization, they represent a significant capability surface if the agent were to be compromised.

company-brainify