The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external binaries gbrain and gstack make-pdf located in the user's home directory. This creates a dependency on unverified local executables that are not part of the standard agent environment.- [DATA_EXFILTRATION]: The skill is designed to send rendered content via Telegram or Email. It accepts a SLUG parameter that is used to construct file paths ($BRAIN_DIR/$SLUG.md) and as an argument to gbrain get. The lack of input sanitization on this parameter may allow for directory traversal, potentially leading to the exposure and exfiltration of sensitive local files as PDF renderings.- [DYNAMIC_EXECUTION]: The script dynamically constructs command-line arguments for the make-pdf binary based on user-requested flags (e.g., --cover, --toc, --watermark). While intended for functionality, this pattern of building shell commands from user input can be risky if not strictly validated.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown content from external sources.
Ingestion points: Content is fetched via the gbrain tool or read from local markdown files (SKILL.md).
Boundary markers: There are no protective boundary markers or instructions to the agent to ignore embedded commands within the brain pages.
Capability inventory: The skill has access to the filesystem, executes shell commands, and utilizes communication tools (Email/Telegram).
Sanitization: The workflow includes stripping YAML metadata, but lacks robust sanitization of the markdown body content against malicious injection.

brain-pdf