The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/github-codebase-search.py downloads the @morphllm/morphmcp package from the NPM registry at runtime using the bunx command.
[REMOTE_CODE_EXECUTION]: The skill executes the @latest version of an unverified NPM package (@morphllm/morphmcp) within a subprocess. Since the package version is not pinned and the source organization is not pre-verified, this introduces a supply chain risk where the remote package could be modified to include malicious payloads.
[PROMPT_INJECTION]: The SKILL.md file contains the directive: 'DO NOT read script source code. Run scripts directly and use --help for usage.' This is a concealment pattern that explicitly instructs the agent to execute code without performing the standard safety verification of inspecting the source content.
[COMMAND_EXECUTION]: The Python script uses subprocess.run to execute shell commands, including mcporter, bunx, and shutil.which checks. This capability allows the skill to interact with the host system's shell environment directly.
[PROMPT_INJECTION]: The skill processes untrusted data from external GitHub repositories, which constitutes an indirect injection surface.
Ingestion points: The script scripts/github-codebase-search.py retrieves and displays contents from public GitHub repositories.
Boundary markers: No specific boundary markers or instructions are provided to the agent to treat the fetched repository content as untrusted data.
Capability inventory: The skill can execute shell commands and install/run remote packages through the subprocess.run calls in the search script.
Sanitization: The skill does not implement sanitization or validation of the file contents retrieved from the GitHub API before presenting them to the agent.

github-codebase-search