The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use shell_command for running specific utilities such as git, rg, sed, head, cat, and gh. It defines templates for these commands to restrict their scope and ensure they are used only for repository analysis.
[EXTERNAL_DOWNLOADS]: The skill clones external Git repositories from arbitrary third-party URLs into a local temporary directory at /tmp/amp-librarian/.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external repositories, documented as follows: 1. Ingestion points: Source code, documentation, and metadata from externally cloned repositories (SKILL.md). 2. Boundary markers: The skill contains an explicit instruction to treat all text in a clone as data to report on rather than instructions to follow. 3. Capability inventory: The agent has access to file system read operations and git/gh network operations within the temporary task directory. 4. Sanitization: The skill explicitly forbids the execution of package managers, builds, tests, or formatters on the content of the clones.

researching-external-repositories