The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/check_doc_commands.py tool identifies and executes shell commands extracted from Markdown code blocks via subprocess.run. This is performed to validate that documented examples produce the expected output.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to run verification scripts on repository documentation, which can lead to the execution of malicious commands if the source files are untrusted.
Ingestion points: Markdown files such as README.md and CONTRIBUTING.md processed by scripts/check_doc_commands.py.
Boundary markers: The skill does not employ delimiters or specific instructions to ignore malicious embedded commands within the files being verified.
Capability inventory: Shell command execution via subprocess.run within the verification script.
Sanitization: No sanitization or safety checks are applied to commands before they are executed.
[DYNAMIC_EXECUTION]: The scripts/check_doc_commands.py script implements dynamic execution by parsing and running code blocks found within documentation at runtime.
[EXTERNAL_DOWNLOADS]: The assets/verify.sh script provides the option to use bunx to run markdownlint-cli2, which involves fetching the linter tool from the official npm registry.

document-codebases