The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): Several enforcement scripts (e.g., require-commit-before-tested.py, require-dependencies.py) utilize subprocess.run to check the local environment, such as repository status via git or running test suites via pytest. These are legitimate uses for quality assurance and environment validation.\n- [EXTERNAL_DOWNLOADS] (LOW): The skill's documentation references external tools and packages, specifically @anthropic-ai/sandbox-runtime and anthropic-experimental/sandbox-runtime. While these organizations are not on the strict whitelist provided in the instructions, they are recognized official security tools for sandboxing. The skill does not automate their installation, only providing them as references for environment hardening.\n- [PROMPT_INJECTION] (SAFE): No malicious instructions or patterns designed to bypass system safety protocols or override agent instructions were detected. The skill is designed to implement constraints, not remove them.\n- [DATA_EXFILTRATION] (SAFE): No patterns of unauthorized data access or external data transmission were identified. The scripts strictly monitor local tool inputs and state files (state.json, feature-list.json).

enforcement