The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured data from files in the workspace to make automated gating decisions, creating an attack surface for indirect instruction injection.
Ingestion points: JSON summary files are read by gate_ptq.py, gate_run.py, gate_compare.py, and gate_verbosity.py to determine workflow progress.
Boundary markers: The skill uses structured JSON parsing but does not implement explicit delimiters or instruction-bypass warnings for the content processed from these files.
Capability inventory: The skill executes shell commands to run gating scripts and coordinates multiple other domain skills including quantization and deployment.
Sanitization: Scripts include basic validation for data types (numeric scores) and terminal statuses (e.g., SUCCESS), but do not filter for natural language instructions embedded within metadata or error fields.
[DYNAMIC_EXECUTION]: The skill's testing infrastructure employs dynamic module loading to initiate its test framework.
Evidence: The tests/test_gates.py script uses the __import__ function to dynamically load and execute the pytest runner at runtime.

day0-release