The Agent Skills Directory

[COMMAND_EXECUTION]: The file evals/run-static-checks.sh is a shell script designed to verify the skill's structure and integrity. It executes a Python validator (check-skill-static.py) located in the parent repository's script directory. This is standard behavior for testing and maintaining skill quality within a development environment.
[DATA_EXFILTRATION]: The skill's templates/optimize-plan.md instructions specify creating tracking artifacts in docs/audits/ or audit-artifacts/. These are local file-system writes intended for documenting optimization progress and do not involve unauthorized network transmission of sensitive data.
[PROMPT_INJECTION]: The skill implements robust boundary definitions and negative test cases (found in evals/activation-cases.md and evals/trigger-evals.json). These are designed to ensure the agent correctly routes tasks and refuses to handle out-of-scope requests, such as auditing live systems or refactoring unrelated code, which helps prevent accidental or adversarial behavior shifts.
[SAFE]: The skill follows a well-defined workflow, uses local grounding sources (inspired by established engineering literature), and maintains a clear separation between its design functions and other system-altering capabilities.

perf-design