The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text describing UI artifacts, which could potentially contain malicious instructions intended to influence the agent's judgment. Ingestion points: The artifact text provided by the user and processed in scripts/checks.py. Boundary markers: The skill employs a structured 4-pass protocol (Inventory, Sweep, Severity, Assembly) to maintain focus, although the input format does not use explicit instruction-ignore delimiters. Capability inventory: The skill utilizes local file writing and execution of its own analysis scripts (checks.py, merge.py). Sanitization: Input is parsed by scripts/checks.py using regex and a fixed set of field labels (e.g., 'Message:', 'Placement:').
[EXTERNAL_DOWNLOADS]: The skill instructions require the installation of the jsonschema package from the Python Package Index (PyPI). This is a standard and well-known library used for data validation.
[COMMAND_EXECUTION]: The skill protocol involves executing its own local Python scripts to perform analysis and merge results. This is the intended behavior of the skill and uses local resources.

critique-microcopy