The Agent Skills Directory

[PROMPT_INJECTION]: The file 'references/evals.yaml' contains test cases designed to simulate prompt injection attacks, such as instructions to 'Ignore previous instructions.' These strings are part of a security evaluation suite used to test the agent's resilience and are not active instructions that influence the agent's runtime behavior.
[COMMAND_EXECUTION]: The 'scripts/install-plugin-from-github.py' script executes system commands including 'git' and 'uv' to fetch and validate plugin packages. These operations are well-guarded by multiple security controls, such as a repository allowlist, mandatory pinned commit SHAs, and path traversal protections.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to 'github.com' and 'codeload.github.com' to retrieve plugin source code. These actions target well-known services and are governed by user-defined trust policies and automated provenance verification steps.

plugin-installer