The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted user input (descriptions of roadblocks and attempts) and archiving it verbatim to the filesystem without sanitization. If the agent later processes these logs, it could be influenced by malicious instructions embedded in the historical data. 1. Ingestion points: User input fields for the goal, the wall, and previous attempts in SKILL.md. 2. Boundary markers: Content is stored within a markdown template but lacks explicit instructions for the agent to ignore embedded commands. 3. Capability inventory: Filesystem write access to the user's home configuration directory (~/.config/makerskills). 4. Sanitization: No sanitization or escaping of external content is performed before writing to the archive files.
[COMMAND_EXECUTION]: The skill instructs the agent to create directories and write files to the user's home directory (~/.config/makerskills/unstuck/archive/). While this is a legitimate method for maintaining session history, users should be aware that their problem descriptions and agent reasoning are stored as persistent local files.

unstuck