The Agent Skills Directory

[COMMAND_EXECUTION]: The script discover_environment.py uses subprocess calls to execute standard system utilities like git, cmake, and g++ solely to verify their presence and version information. Similar self-test functions are present in other utility scripts to ensure local functionality before use.
[DATA_EXPOSURE]: The skill implements a strictly local state management and evidence logging system. It focuses on project-specific metadata and does not attempt to access sensitive user files, private keys, or credentials.
[DYNAMIC_EXECUTION]: The environment discovery tool can dynamically extend its tool list via local profile configurations stored in JSON files. This mechanism is restricted to the skill's own directory structure and does not involve remote code loading.
[PROMPT_INJECTION]: The skill's instructions are designed to harden the agent's behavior by prioritizing objective evidence and structured validation over the agent's internal context or memory, which serves as a safeguard against common hallucination and drift issues.

long-horizon-skills