The Agent Skills Directory

[COMMAND_EXECUTION]: The testing file scripts/tests/test_edge_cases.py uses subprocess.run to execute the skill's utility scripts (context_analyzer.py and compression_evaluator.py) as part of its automated validation suite. This is a standard practice for functional testing of CLI tools.
[DATA_EXPOSURE]: The reference document references/runtime-awareness.md mentions configuration file paths such as ~/.claude/.credentials.json and %USERPROFILE%\.claude\.credentials.json. These are documented as part of an explanation on how the host platform (Claude Code) manages authentication and usage monitoring. The provided Python scripts do not attempt to access or exfiltrate these files.
[PROMPT_INJECTION]: The skill features scripts that process potentially untrusted external data (JSON-formatted conversation histories).
Ingestion points: scripts/context_analyzer.py (via load_json_file) and scripts/compression_evaluator.py (via load_file).
Boundary markers: Not explicitly defined in scripts.
Capability inventory: File read/write via scripts; subprocess execution in tests.
Sanitization: Standard JSON parsing is used, with a 100MB file size limit implemented to prevent resource exhaustion.

context-engineering