eval-runlisted

# Eval Run Run ID: `$ARGUMENTS` If no run ID was provided, stop and show: ```text /eval-run <run-id> ``` Set `<run-dir>` to `evals/runs/$ARGUMENTS`. ## Isolation rule The run must measure discovery, not reproduction. During this skill: - read only the active run's `case.yaml`, its own artifacts, relevant framework files, agents, and normal Venture OS skills; - never read `evals/reference/`; - never inspect git history, prior eval runs, or other Case Library entries for expected answers; - external market research is allowed; - if reference expectations are already present in context, stop and report the run as contaminated instead of continuing. ## Workflow 1. Read `<run-dir>/case.yaml` and `<run-dir>/metadata.json`. 2. Use `<run-dir>/venture/` as the target venture directory. 3. Apply `venture-new` from the seed idea without doing research in that step. 4. Apply `venture-research` against the critical assumptions. 5. Apply `venture-challenge` adversarially. 6. Apply `venture-decide` and persist the gate decision. 7. If the decision's `next_action` is an experiment (`type: experiment`), apply `venture-experiment` for the assumption it targets and run `npm run experiment:lock -- <path-to-experiment.yaml>`. Leave the experiment `designed`: the run ends before anything is executed. Freeze refuses a run whose decision chose an experiment that was never designed and locked with routed success and failure outcomes, because a later result could not be checked against it. 8