The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided CSV return series and JSON configuration files, presenting a potential surface for indirect prompt injection.\n
Ingestion points: scripts/analyze_residual_edge.py reads data from paths specified in the --input and --config arguments.\n
Boundary markers: The script implements strict deterministic validation, rejecting malformed dates or non-finite numbers. Instructions mandate manual interpretation of results.\n
Capability inventory: The skill performs file reads, mathematical modeling, and file writes. It does not possess network access or arbitrary command execution capabilities.\n
Sanitization: Input is parsed through json.loads and csv.DictReader with explicit type checks and schema version enforcement.\n- [COMMAND_EXECUTION]: The testing suite utilizes shell command execution to perform integration tests on the CLI.\n
Evidence: scripts/tests/test_analyze_residual_edge.py calls subprocess.run() to execute the analysis script using sys.executable. This is a routine and safe testing practice for verifying command-line tools.\n- [DYNAMIC_EXECUTION]: The skill uses dynamic loading within its test configuration.\n
Evidence: scripts/tests/conftest.py employs importlib.util to load the local analysis script as a module for testing. This is a standard testing pattern that does not involve untrusted remote code.

residual-edge-analyzer