The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to extract and execute commands directly from repository-controlled configuration files without verification.
Ingestion points: Commands and environment variables are read from .github/workflows/*.yml and dependency versions from package-lock.json.
Boundary markers: The instructions lack boundary markers or warnings, explicitly advising to copy and run logic "verbatim".
Capability inventory: The agent utilizes shell command execution, file system access (writing .env.ci), and network access via the GitHub CLI.
Sanitization: No sanitization or filtering is performed on the extracted command strings before local execution.
[COMMAND_EXECUTION]: Extensive use of shell utilities and development tools for environment setup and diagnostics.
Utilizes sed and grep for parsing configuration files.
Runs test frameworks and linters such as pytest and ruff.
Manages virtual environments and toolchains via uv and npm.
[EXTERNAL_DOWNLOADS]: The skill recommends downloading specific versions of tools using established package managers.
Fetches the ruff linter using uvx from PyPI.
Fetches the prettier formatter using npx from the npm registry.

reproducing-ci-locally