The Agent Skills Directory

[COMMAND_EXECUTION]: The deployment script scripts/automation/deploy.sh uses chmod +x to grant execution permissions to diagnostic and analysis scripts within the skill's own directory, subsequently running validation tests.
[COMMAND_EXECUTION]: Documentation for integration testing (modules/integration-testing.md) and performance profiling (modules/performance-benchmarking.md) provides Python implementation examples that utilize subprocess.run to execute local tools to measure execution latency and functional correctness.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of other skill files, presenting an Indirect Prompt Injection surface (Category 8). Ingestion points: The skill reads SKILL.md and auxiliary module files from the local filesystem (~/.claude/skills/). Boundary markers: The framework explicitly implements pressure-testing.md and anti-rationalization.md modules designed to identify and resist adversarial instructions embedded in audited data. Capability inventory: The skill performs local filesystem read operations and executes internal analysis scripts via subprocess. Sanitization: It utilizes a structural compliance-checker to validate that audited files adhere to metadata schemas and safety standards.

skills-eval