The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/test_skill.py uses subprocess.run to execute the skill's internal validation tools (validate.py and test_validate_agents_md.py). These executions are performed using the active Python interpreter (sys.executable) and target local script files within the skill package, which is a standard implementation for automated testing suites.
[REMOTE_CODE_EXECUTION]: The regression test script scripts/test_validate_agents_md.py utilizes importlib.util to dynamically load the validate_agents_md.py logic. This is a local operation used to facilitate testing of the validator against temporary directory fixtures and does not involve remote code retrieval or arbitrary command execution.
[PROMPT_INJECTION]: The skill instructions in SKILL.md implement a defensive strategy against indirect prompt injection by mandating that existing instruction files be treated exclusively as evidence. The agent is explicitly told to disregard the wording, structure, and assumptions of pre-existing AGENTS.md or CLAUDE.md files, ensuring that potentially malicious or stale instructions are not inherited during the replacement process.
[EXTERNAL_DOWNLOADS]: The references/source-notes.md file contains several URLs pointing to official developer documentation from well-known service providers including OpenAI, Google, and Anthropic. These function as citations for the skill's design principles and do not trigger any runtime downloads or execution of external payloads.

bootstrap-agents-md