The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/optimize_description.py script executes the claude CLI tool using subprocess.run to automate the evaluation of skill descriptions. The command is invoked using an argument list rather than a shell string, which is a secure implementation.
[DYNAMIC_EXECUTION]: The script scripts/smoke_test_scoring.py generates a temporary Python file at runtime and marks it as executable to mock the claude tool's behavior. This is used for deterministic unit testing of the scoring logic and is confined to the test environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process the content of other skills found in the local ~/.claude/skills directory. This creates a potential surface where adversarial instructions in audited files could influence the agent's behavior. To mitigate this, the skill uses XML-style boundary markers (e.g., <skill_content>) when interpolating these files into prompts.

yzr-skill-creator