Lucasuiii/modeling-workbench / .agents/skills/cumcm-workflow/assets/independent-review
cumcm-independent-review
VerifiedIndependently review a CUMCM computation package before its claims enter validation or paper writing. Use only inside a generated independent-review package.
CUMCM Independent Review
Review this freshness-bound package without consulting the originating conversation. Treat conclusions as untrusted and reconstruct only what is needed from official inputs, contracts, selected source, official runs, and outputs.
Boundaries
- Work read-only inside the package.
- Do not search for missing official materials. Report them as missing.
- Do not edit, rerun, or replace the preserved execution unless the user separately authorizes a reproduction run.
- File existence and successful execution do not prove that the model answers the official question.
- Verify the package/upstream bindings before substantive review; a stale package is inconclusive.
- Give exact file, formula, code, or numerical locations for every P0/P1 finding.
- Preserve negative and inconclusive findings verbatim.
Review order
- Read
REVIEW_REQUEST.mdandmaterials/problem/SOURCE_MANIFEST.json. In targeted mode, also read package-rootTARGETED_FINDINGS.json; it is the self-contained prior-P0 brief, so do not request the complete prior review. - Reconstruct each subproblem from the supplied official files and
PROBLEM_FACTS.json. - Compare the official request with the model objective, variables, constraints, assumptions, and cross-question dependencies.
- Inspect computation entry points, run manifests, executed outputs, and result locators.
- Challenge relevant failure classes:
- task or target misunderstood;
- upper/lower bound or optimization direction reversed;
- a quantity counted twice;
- unsupported extrapolation;
- an observed variable omitted without justification;
- cross-question contradiction;
- code and mathematical formulation disagree;
- numerical output violates units, bounds, conservation, or official constraints.
- For targeted mode, resolve every entry in
TARGETED_FINDINGS.jsonfirst. Do not repeat a full review unless the target change has global impact or current evidence reveals a new, well-supported P0. - Write the raw review and structured result using the supplied template.
Verdict
accepted: no open P0 and no material unresolved concern in scope.accepted_with_concerns: no open P0; one or more P1 concerns remain.revision_required: at least one open P0 requires returning to the earliest affected stage.inconclusive: required material is missing or the available evidence cannot support a decision.
Classify findings as P0/P1/P2 and open/resolved/accepted_concern. State the reviewer, model if applicable, originating/reviewer task references, and independence grade. Never describe same-context review as independent; a same-model fresh task remains correlated.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |