The Agent Skills Directory

[SAFE]: No malicious patterns, such as credential theft, prompt injection, or unauthorized remote code execution, were detected. The skill functions as a local documentation and scripting aid for code quality workflows.
[COMMAND_EXECUTION]: The script scripts/test_skill.py utilizes subprocess.run to execute the local helper script scripts/classify_conversion.py during self-verification tests. The command uses a list of arguments and sys.executable without a shell, which is the secure implementation of subprocess execution. The deterministic detector's flag for shell execution is a false positive in this context.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow to codify session learnings into executable artifacts. While this involves processing natural language heuristics, the risk is mitigated by the skill's focus on developer-led structural verification and existing toolchain reuse.
Ingestion points: Natural language heuristics provided during the agent workflow in SKILL.md.
Boundary markers: Instructions guide the separation of learnings from enforcement, though no formal technical delimiters are used for raw input.
Capability inventory: Directs agents to create and modify local validation scripts, Git hooks, and CI configurations.
Sanitization: Guidelines emphasize utilizing repository-standard JSON schemas, formatters, and existing test suites for enforcement.

heuristic-to-deterministic