mock-review
Skill15 starsChanged 10 days agoMITView on GitHub
Use before submission (about 3 days and again 1 day before) or before a rebuttal to simulate venue-calibrated reviews. Runs falsify-first reviewer personas (expert-skeptic, generalist, implementer, plus optional data-auditor and desk-reject auditor), verifies their findings, and outputs scored reviews plus an executable fix list.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- ANT1OFF/gramtheir own version, first seen Nov 8, 2026
- Immortalqx/my_codex_skillstheir own version, first seen Jun 10, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- Immortalqx/my_codex_skills★ 54same name
- ANT1OFF/gram★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository29 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- abstract-craftSkill
- anonymize-checkSkill
- artifact-packSkill
- baseline-pinSkill
- bench-protocolSkill
- bib-verifySkill
- camera-readySkill
- eprint-searchSkill
- experiments-writingSkill
- falsifySkill
- figure-harvestSkill
- idea-mining-loopSkill
- lean-bridgeSkill
- lit-matrixSkill
- log-to-evidenceSkill
- paper-figuresSkill
- paper-playbookSkill
- param-estimationSkill
- phase-gateSkill
- polish-writingSkill
- reading-notesSkill
- rebuttalSkill
- related-work-writingSkill
- sage-checkSkill
- technique-overviewSkill
- textbook-indexSkill
- venue-calibrationSkill
The file
Files
Mock review
Principles
- Falsify first. The writer first produces a list of falsifiable statements
(from
CLAIMS.mdand the contribution list: "the packing bound k ≤ N/(2tw) is tight", "the distinguisher holds for every key", "hypothesis H is necessary"). Reviewers are told to break them, not to praise. - Personas are views of one analysis, not independent experts. Say so in the output. Do not present simulated scores as a prediction of the real outcome.
- Only findings that can be located in the source, data or a script are accepted. Reviewer agents produce false positives; the coordinating agent verifies each finding (reproduce the counterexample, open the log, recompute the cell).
- Read-only. Mock reviewers never edit the paper; they write
REVIEWS.md. - Evidence-based and constructive. Each weakness cites a page/section/equation and proposes a fix.
Personas
| persona | stance | checks |
| Expert-skeptic (theory) | senior expert in the exact subfield, assumes the result is wrong | counterexamples on small parameters; hypotheses vs the paper's own parameter table; sufficient-vs-necessary confusions; body vs appendix formula mismatches; proof gaps; overclaimed proof status (heuristic stated as theorem) |
| Generalist (novelty/writing) | good cryptographer outside the subfield | can I restate the idea after the overview? is the delta over the closest prior work explicit? is "unified"/"first"/"optimal" justified? is the body self-contained without the appendix? |
| Implementer (experiments) | builds the same thing in practice | baseline strength/config; same machine/session; scope of ranges; excluded instances; overheads disclosed; ablation present; artifact runs; security estimates per instance |
| Data auditor (optional) | accountant | recomputes every ratio, sum and derived parameter in every table from EVIDENCE rows and logs |
| Desk-reject auditor (optional) | PC chair | page limit, template, fonts, page numbers, anonymity (run anonymize-check), links resolve, references complete |
| Links/anonymity (optional) | curl every URL; grep identity strings in PDF text and binary; artifact README wording |
Pick 3 (default) to 5 personas; weights and venue anchors from venue-calibration.
Procedure
- Brief (
reviews/REVIEW-BRIEF.md): venue, paper path + hash, falsifiable statement list, persona assignments, what is out of scope (e.g. "we do not rerun the full benchmark; historical logs are not presented as fresh reproduction"). - Dispatch persona agents in parallel, each with the brief and instructions to try to refute first.
- Verify each finding: reproduce counterexamples with a script, recompute numbers,
open cited logs. Mark each finding
confirmed | false-positive | unverifiable. - Score with the six-point scale + confidence 1–4 per persona, plus dimension scores (problem value, novelty, technical soundness, experimental persuasiveness, precision of statements, reproducibility). Dimension scores are not averaged into an acceptance probability.
- Fix list (
reviews/SUBMIT-FIXES.md): items graded must-fix (compliance, integrity, confirmed errors) / should-fix / won't-fix (with reason), each with file:line, old text, new text, and why. Include "good news" items (what was checked and found correct) so authors do not second-guess sound parts. - Rebuttal prep: list the 3–6 questions most likely to be asked and the evidence
that answers each (feeds the
rebuttalskill). - Human author decides which fixes to apply; after fixes, rerun the implementer (data) and desk-reject personas.
Output format (REVIEWS.md)
## Round <k> — <date> — venue profile: <venue> — paper hash <sha256[:12]>
| persona | score | confidence | top-3 weaknesses | action item → agent |
|---|---|---|---|---|
Scope and limits of this review: <what was read, what was not rerun>
Personas are views of one analysis, not independent experts.
## Reviewer A — <persona>
Summary: 2–3 sentences.
Strengths: S1 ... S2 ... S3 ...
Weaknesses: W1 (severity high|med|low, location, evidence, suggested fix) ...
Questions: Q1 ... Q2 ...
Scores: overall x/6, confidence y/4; dimensions: ...
Recommendation: ...
## Reviewer B ...
## Reviewer C ...
## Meta-review (simulated discussion)
Score table | consensus points | disagreements | decisive issues |
subjective band with conditions (not a statistic) | top preparation items for rebuttal
## Verified findings
| id | finding | status confirmed/false-positive | evidence (script/log) |Frequent catches (generic)
| category | catch | prevention |
| anonymity | footnote URL to a personal code host account | anonymous mirror only; run anonymize-check |
| theory | body formula differs from appendix formula | copy from the proof source, never retype |
| theory | unconditional statement with a small counterexample | exhaustive search on toy parameters |
| theory | theorem hypothesis excludes some of the paper's own parameter sets | run the parameter table through each theorem |
| data | headline scope includes instances below the claimed security level | generate scope sentences from the security table |
| data | two incommensurable counters in one column | state each column's counter source in the caption |
| writing | composite gain credited to one component | label each number with its configuration |
| writing | "rigorous guarantee" for an asymptotic/heuristic result | "asymptotic, certified per instance" |
| citations | fabricated or wrong metadata, wrong author count | lit-scout verifies each entry against DOI/ePrint |
| layout | overprinting after negative vspace near floats | PDF overlap scan |
| strategy | best evidence (an ablation) not in the paper | ask "which experiment would convince a skeptic?" and check it is in the body |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.