The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes Python scripts using python3 -c by interpolating RESOLVED_PATH directly into single-quoted string literals. If a resolved file path contains a single quote, it could break the script syntax or lead to arbitrary code execution within the Python environment.
Evidence: info = geoai.get_raster_info('RESOLVED_PATH') inside the python3 -c blocks in Step 3 and Step 5.
[PROMPT_INJECTION]: User-provided input from the question argument ($1) is interpolated directly into the agent's instructional context without sanitization or boundary markers.
Evidence: ${1:-describe the data: summarize the file type, CRS, extent, and any notable properties.} in Step 4.
Risk: An attacker could provide a 'question' that contains instructions designed to override the agent's primary goals or safety filters.
[COMMAND_EXECUTION]: The skill creates and modifies a state file in the user's home directory, which constitutes a persistence mechanism.
Evidence: STATE_DIR="$HOME/.geoai-skills/$PROJECT_ID" and subsequent json.dump operations to state.json.
[SAFE]: The skill utilizes the geoai Python library. As the skill is authored by opengeos, this is identified as a legitimate vendor resource.

inspect-geo