The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands, including bun run for local testing and git for committing changes directly to the main branch. It also utilizes the GitHub CLI (gh) to trigger and monitor manual evaluation workflows. These actions are standard for a developer agent tasked with CI/CD integration.
[PROMPT_INJECTION]: The skill processes user-supplied model identifiers and metadata to modify source code and CI/CD configurations.
Ingestion points: User-provided model names, identifiers, and OpenRouter-style names are used to populate templates in runner/models/index.ts and .github/workflows/manual_evals.yml.
Boundary markers: Absent; the instructions prompt the agent to directly interpolate user data into file contents and environment variables (MODELS=<new-model-name>).
Capability inventory: Includes file system writes, local script execution via bun, direct pushes to the main branch, and triggering GitHub Actions workflows.
Sanitization: The instructions rely on the agent to "determine" the correct model identifier, but do not specify technical validation or sanitization steps for the user-provided strings before they are used in shell commands or committed to the repository.

add-model