buildinternet/releases

buildinternet/releases

Agent Claude Code

Generates an org "overview" knowledge-page body from a composed prompt (the production overview system prompt + release inputs). Used by the key-free sub-agent overview eval (tests/evals/overview-subagent flow) to exercise overview generation on the Claude Code subscription instead of the metered Anthropic API.…

buildinternet/releases

Agent Claude Code

Generic rubric grader for the local key-free evals. Reads a grader prompt (a + an plus grading instructions, as produced by buildGraderPrompt) and returns a single JSON verdict {criteria, explanation, result}. The grading logic is domain-neutral — the rubric supplies the per-eval criteria — so one grader serves every…