Borrowing it
Nothing to install: this file belongs to htlin222/robust-lit-review. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/htlin222/robust-lit-review/main/.claude/skills/claim-appraise/SKILL.mdgit clone --depth 1 https://github.com/htlin222/robust-lit-reviewWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/htlin222/robust-lit-review/claim-appraise)<a href="https://agentmods.dev/skills/htlin222/robust-lit-review/claim-appraise"><img src="https://agentmods.dev/badge/skills/htlin222/robust-lit-review/claim-appraise/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/htlin222/robust-lit-review/claim-appraise"><img src="https://agentmods.dev/badge/skills/htlin222/robust-lit-review/claim-appraise.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00069 | $0.01357 |
| Opus 5 | $0.00034 | $0.00678 |
| Sonnet 5 | $0.00014 | $0.00271 |
| Haiku 4.5 | $0.00007 | $0.00136 |
Grade A, and why
claim-appraise scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Claim Appraisal Workflow (/claim-appraise)
You are turning ONE real-world claim into a defensible, peer-checkable evidence
appraisal. Unlike /lit-review (which reviews a topic into a manuscript), this
adjudicates a claim and renders a verdict per sub-question.
Worked example shipped with this repo: the 4+2R 代謝飲食法 appraisal → https://verdict-4plus2r-sr.pages.dev/.
When to use
"Is legit?", "appraise the diet/supplement/protocol", "fact-check this health claim".
Hard quality gates (non-negotiable)
- Literature ≥ 2016 and Q1 journals only (
assess_journal_quality(strict=True)+filter_by_year). - Every DOI resolved via doi.org and CrossRef-verified to exist (
utils/crossref.py) — the anti-hallucination gate. Never cite a work CrossRef can't confirm. - Per sub-question GRADE certainty (recomputed deterministically, LLM rating is advisory).
- OpenEvidence cross-check where the relay is available (non-fatal if not).
- Overall verdict logic audited with
/argdownbefore wording ships. - Attach the claim's primary source URL so peers can confirm no 斷章取義 (cherry-picking).
Steps
-
Identify the real claims. Read the claim's PRIMARY source (official site / book), not secondary blogs. Enumerate every distinct sub-claim. Save the source URL(s).
-
Decompose into PICO sub-questions — one per distinct outcome domain (efficacy AND safety AND any over-extended claims). See
scripts/run_4plus2r.pyfor thePICOQuestionshape. Get the list approved before searching (wrong PICOs waste the run). -
Run the per-PICO SR (real API search; needs
.envkeys). For each PICO:ClaimAppraisalPipeline.run_pico_search→ Scopus+PubMed+Embase → dedup → year≥2016 → Q1-strict → DOI-validate → CrossRef-verify →PicoPrismaFlow+ included studies. Drive it likescripts/run_4plus2r.py(savesoutput/<claim>/pico_NN.json). -
Write one chapter per PICO following
output/4plus2r/CHAPTER_SPEC.md: dispatch a subagent per chapter that reads only itspico_NN.json, selects the most relevant HUMAN studies (flag animal/in-vitro as 機轉假說), assigns a verdict chip (vc-strong/moderate/weak/contra/risk/unproven) and GRADE pips, and cites via<cite class="ref" data-doi="DOI">. NEVER invent a citation.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 67 lines · 69 tokens per session scan A dd29f62d7c12
claim-appraise is a skill published in the GitHub repository htlin222/robust-lit-review (56 stars, last pushed 3d ago), licensed MIT. It adds 69 tokens to every session and 1,357 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
cirq
Google quantum computing framework. Use when targeting Google Quantum AI hardware, designing noise-aware circuits, or running quantum characterization experiments. Best for Google hardware, noise modeling, and low-level circuit design. For IBM hardware use qiskit; for quantum ML with autodiff use pennylane; for…
qiskit
IBM quantum computing framework. Use when targeting IBM Quantum hardware, working with Qiskit Runtime for production workloads, or needing IBM optimization tools. Best for IBM hardware execution, quantum error mitigation, and enterprise quantum computing. For Google hardware use cirq; for gradient-based quantum ML use…
geopandas-spatial
Geospatial and climate data analysis with geopandas and xarray. Use when: (1) geographic vector data analysis, (2) climate and weather NetCDF data, (3) spatial joins and overlay operations, (4) map visualization, (5) CRS transformations and area calculations. NOT for: satellite imagery ML (use rasterio/torchgeo)…
pymatgen-materials
Materials science computation with pymatgen. Use when: (1) crystal structure creation and manipulation, (2) phase diagram construction, (3) electronic structure analysis, (4) symmetry and space group operations, (5) VASP input/output parsing. NOT for: molecular chemistry (use rdkit-chemistry), protein structure (use…
rdkit-chemistry
Molecular chemistry operations via RDKit. Use when: user asks about molecular structures, SMILES, chemical properties, or fingerprints. NOT for: reaction databases or wet lab protocols.
research-agent
Project-first Research pipeline with live gates, source/PDF evidence, claim boundaries, compute and publication controls; includes an optional PubMed/arXiv standard-library helper.