The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts (check_patch_scope.py, merge_audit_reports.py) and unit tests (test_scripts.py) that invoke git via the subprocess module to inspect repository state, calculate line churn, and initialize test environments. These operations are essential for the skill's purpose but involve shell command execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data, including repository source code, git diffs, and structured JSON audit reports. This creates a surface for indirect prompt injection if the processed content contains malicious instructions intended to influence the agent's behavior during the review or change phases. 1. Ingestion points: Source files and diffs via git commands in check_patch_scope.py; JSON report files in merge_audit_reports.py; full repository scanning in the AUDIT and SWEEP states. 2. Boundary markers: The skill instructions specify that quoted commands remain data and that prose/soul modes do not override core guidance rules. It uses structured JSON for report merging to maintain data integrity. 3. Capability inventory: File system read access; local command execution via git subprocesses; authorized file write capabilities during the CHANGE state. 4. Sanitization: merge_audit_reports.py performs normalization of file paths and text content (whitespace collapsing and case-folding) to ensure deterministic merging.
[DYNAMIC_EXECUTION]: The tests/test_scripts.py file uses importlib.util to dynamically load the skill's utility scripts for verification. While this is used for internal testing purposes and targets local files within the skill's own directory, it represents a dynamic code loading pattern.

code-quality-workflow