The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text instructions which are interpolated into a 'meta-prompt' sent to an LLM to generate annotation instructions.
Ingestion points: The instruction argument provided to the scripts/annotate.py script.
Boundary markers: The script uses clear textual delimiters such as ANNOTATION REQUEST: and CRITICAL: instructions in its internal prompt templates to isolate user input.
Capability inventory: The script has the ability to read image files, write to the file system (outputting annotated images), and access the network specifically via the Google Generative AI client.
Sanitization: The input is joined as a string; while no explicit regex-based sanitization is present, the impact is limited to the visual annotation of the provided screenshot.
[COMMAND_EXECUTION]: The skill defines a command-line interface for its operation but does not use dangerous functions like os.system or eval to process user input or execute arbitrary shell commands.

screenshot-annotator