The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell templates to locate and run a bundled Python script (grill_log.py) for logging session data. The discovery process involves standard tools like git and find to locate the script within the workspace or agent skill directories.
[PROMPT_INJECTION]: The skill ingests user input and writes it to Markdown logs, which are subsequently read by the agent for session context. This creates a surface for indirect prompt injection (Category 8).
Ingestion points: User answers are processed through scripts/grill_log.py via the --answer and --resolved-answer arguments and stored in workspace files (e.g., tmp/grill-me/session-*.md).
Boundary markers: All logged user content is wrapped in Markdown quote blocks (> ) to distinguish it from the agent's instructions and structure.
Capability inventory: The agent has the capability to read and write files and execute shell commands within the workspace directory.
Sanitization: Inputs are blockquoted to prevent formatting breaks, and session keys are sanitized of non-alphanumeric characters. No complex sanitization of the answer content is performed before logging.

grill-me