adversarial-review
Skill4 starsChanged 20 days agoMITView on GitHub
Run an adversarial peer-review swarm on a paper draft, report, or manuscript by spawning parallel persona subagents (hostile theorist, experimentalist, statistician, journal editor, citation auditor, figure critic) and synthesizing their critiques into a ranked fix list with suggested defense experiments. Use when the user asks for a paper review, referee simulation, desk-reject check, pre-submission audit, citation audit, figure audit, novelty attack, or wants feedback on a draft report before submission or internal circulation.
109 other projects use a version of this.
Adopted by
Most recently taken up by
- anton-abyzov/specwithcodetheir own version, first seen Nov 16, 2026
- magnusrodseth/dotfilestheir own version, first seen Nov 16, 2026
- lennoxsaint/Skillstheir own version, first seen Nov 15, 2026
- fabioc-aloha/Alex_ACT_ONEtheir own version, first seen Nov 15, 2026
- samuelralak/How-I-Worktheir own version, first seen Nov 15, 2026
- baekenough/oh-my-customcodetheir own version, first seen Nov 14, 2026
Used with
Projects with it also use rethink, root-cause-analysis and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
fabioespindulaSkillrethink
Reframe a plan, idea, or decision before the user commits to it. Use when the user asks to rethink, step back, pressure-test, gut-check, decide, right-size, simplify, think bigger about, or evaluate a concrete plan or decision in product, strategy, technical architecture, career, life, company direction, or personal priorities. This skill improves decision quality before action: clarify the real question, challenge assumptions, compare alternatives, calibrate ambition, classify the operating mode, surface risks, and recommend a next move. Do not use for broad future exploration without a concrete decision; that belongs to a think-big or research-style skill.
Also in 3 projects that use this
MicrosoftSkillroot-cause-analysis
Use when a Regional Retail Manager asks I'm seeing unusual inventory movement at our Northwest stores. Can you help me understand what's happening?
Also in 3 projects that use this
anasahmed07Skillbetter-auth-best-practices
Skill for integrating Better Auth - the comprehensive TypeScript authentication framework.
Also in 4 projects that use this
DockerSkillfix-issue
Fix a single GitHub issue end-to-end: triage, research, write the fix, review, and create a PR. Use when asked to fix an issue: "fix issue 1234", "resolve #500", "create a PR for issue 200".
Also in 3 projects that use this
MicrosoftSkillpull-request
Write concise, focused pull request descriptions for Wassette — at most three sentences on the what and why, with an explicit note on any breaking public-API change and how users adapt. Use when opening or editing a Wassette pull request.
Also in 3 projects that use this
juriwiensSkillcodex-review
Run code review via Codex CLI
Also in 3 projects that use this
- +2 more from microsoft
Other versions109 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 109
- PrefectHQ/fastmcp★ 29ksame name
- jlowin/fastmcp★ 29ksame name
- mengxi-ream/read-frog★ 10ksame name
- epicenter-md/epicenter★ 4.8ksame name
- EpicenterHQ/epicenter★ 4.8ksame name
- basicmachines-co/basic-memory★ 4.6ksame name
- DanMcInerney/architect-loop★ 626same name
- deadlock-mod-manager/deadlock-mod-manager★ 584same name
- boraoztunc/skills★ 473same name
- syahiidkamil/Software-Engineer-AI-Agent-Atlas★ 404same name
99 more not listed
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 repository38 other files it gives its agents.
- academic-cvSkill
- academic-jobsSkill
- academic-proseSkill
- bibtex-genSkill
- chrome-sessionSkill
- clickup-cliSkill
- clickup-todaySkill
- commit-triageSkill
- concept-explainerSkill
- contact-emailSkill
- cover-letterSkill
- dropboxSkill
- handdrawn-schematicSkill
- hep-rumor-millSkill
- journal-club-reviewSkill
- magi-researchSkill
- md2pdf-typoraSkill
- morgenSkill
- overleaf-section-workflowSkill
- overleapSkill
- peer-agent-reviewSkill
- proton-mailSkill
- reference-searchSkill
- remarkable-syncSkill
- research-backupSkill
- research-boardSkill
- research-logSkill
- research-portalSkill
- research-proposalSkill
- research-reportSkill
- research-statementSkill
- scienceplot-pySkill
- vastaiSkill
- wide-slide-illustratorSkill
- workshop-paper-reviewSkill
- xkcd-pySkill
- zai-web-searchSkill
- zoom-summarySkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. 0 are common, 10 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 98 edited versions we read keep these.
10
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill to stress-test a draft before it reaches real reviewers.
- Distinctive: line 19
If the user wants a friendly copy-edit pass, this is the wrong skill — suggest a plain editing request instead.
- Distinctive: line 23
Ask only for what is missing:
- Distinctive: line 31
If nothing is specified, default to:
- Distinctive: line 41
Before spawning any agents:
- Distinctive: line 43
Read the draft once end-to-end to understand the claim structure.
- Distinctive: line 47
Do not let each persona re-read and re-summarize the whole draft from scratch — give them the packet plus the draft path.
- Distinctive: line 58
Default persona set (skip any the draft does not support):
- Distinctive: line 69
Every persona must return a markdown block with this exact structure so the meta-editor can merge them deterministically:
- Very specific: line 90
After all personas return, run the meta-editor pass described in references/meta-editor.md.
- Distinctive: line 129
Never modify the draft. This skill is read-only with respect to the manuscript.
- Very specific: line 130
Never silently skip a persona.
- Very specific: line 132
Never fabricate a citation when checking citations.
- Very specific: line 133
Never push the review outputs.
“Common” is measured: kept by at least 6 in 10 of the 98 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Adversarial Review
Use this skill to stress-test a draft before it reaches real reviewers. It spawns a small swarm of specialized persona subagents in parallel, each attacking the manuscript from one angle, then synthesizes the critiques into a prioritized fix list.
This is the opposite of a friendly read-through. The goal is to surface desk-reject-level problems, unsupported claims, weak statistics, and citation hallucinations before submission — not to reassure the author.
When to use
- Pre-submission audit for a journal (PRX, JCAP, JOSS, etc.) or conference.
- Internal circulation of a research report or technical memo.
- Response drafting for a revise-and-resubmit where the author wants to predict reviewer pushback.
- Any draft the user wants "torn apart" or "reviewed hard."
If the user wants a friendly copy-edit pass, this is the wrong skill — suggest a plain editing request instead.
Inputs to confirm
Ask only for what is missing:
- path to the draft (default:
report.mdin the current directory) - journal or venue the draft targets (affects the Journal Editor persona)
- any sections the user wants emphasized or skipped
- whether figures live alongside the draft and where (
plots/,figures/, etc.) - whether a bibliography file or a references section should be audited
- whether Korean translation of the final summary is expected
If nothing is specified, default to:
- draft:
report.mdin CWD - venue: "general physics/ML journal" (Journal Editor persona stays generic)
- figures:
plots/if present, else skip Figure Critic - Korean translation: no, unless the user previously asked
Core workflow
1. Locate and stage inputs
Before spawning any agents:
- Read the draft once end-to-end to understand the claim structure.
- Enumerate figure paths and bibliography entries.
- Produce a compact context packet that every persona will receive: draft path, venue, figure list, bibliography location, and any sections to emphasize or skip.
Do not let each persona re-read and re-summarize the whole draft from scratch — give them the packet plus the draft path.
2. Spawn personas in parallel
Spawn the persona subagents using the Agent tool, in a single message so they run concurrently. Use references/personas.md for the exact persona briefs; each persona gets:
- the context packet
- its own persona brief (role, what to attack, what to ignore)
- an output contract (see below)
- an explicit length cap so critiques do not bloat
Default persona set (skip any the draft does not support):
- Hostile Theorist — attacks novelty; uses
reference-searchto surface prior art the draft may have missed. - Experimentalist — attacks reproducibility, missing ablations, undisclosed controls, hyperparameter sensitivity.
- Statistician — attacks error bars, significance claims, baseline fairness, multiple-comparison exposure.
- Journal Editor — predicts desk-reject reasons for the target venue; checks scope fit, framing, required sections.
- Citation Auditor — verifies every non-trivial citation resolves to a real paper via
reference-search; flags hallucinations and mischaracterizations. - Figure Critic — checks figure paths resolve, captions match the data, axes are consistent, cutoffs are applied uniformly, and no figure is referenced but missing.
3. Output contract (per persona)
Every persona must return a markdown block with this exact structure so the meta-editor can merge them deterministically:
# <Persona Name>
## Severity-classified findings
- **DESK-REJECT**: <finding> — <evidence pointer: section/figure/citation>
- **MAJOR**: <finding> — <evidence pointer>
- **MINOR**: <finding> — <evidence pointer>
## Suggested defense / fix
- <action the author should take>
## Confidence
<high | medium | low> — <one sentence why>Personas that find nothing at a severity level simply omit that bullet. Never fabricate findings to fill the template.
4. Meta-editor synthesis
After all personas return, run the meta-editor pass described in references/meta-editor.md. It:
- deduplicates overlapping critiques
- ranks by severity (DESK-REJECT > MAJOR > MINOR)
- attaches a proposed defense experiment or textual fix to every DESK-REJECT and MAJOR
- drops low-confidence findings that no other persona corroborates
Meta-editor output is the canonical summary.md the user reads first.
5. Write outputs
Write all artifacts under outputs/review/<YYYYMMDD_HHMM>/ relative to the draft's directory, with the stamp taken from date +%Y%m%d_%H%M. Alphabetical order then equals chronological order across repeated reviews. Keep individual persona files alongside the synthesis so the user can trace any claim in summary.md back to its source.
outputs/review/20260418_1430/
summary.md # meta-editor synthesis — read this first
hostile_theorist.md
experimentalist.md
statistician.md
journal_editor.md
citation_auditor.md
figure_critic.md
context_packet.md # the packet given to every persona (for reproducibility)6. Report and stop
Show the user the path to summary.md, the count of findings by severity, and the three highest-severity items inline. Do not start fixing the draft. The user decides which findings to act on.
Integration
See references/integration.md for how this skill wires into the rest of the collection:
reference-searchis the citation and prior-art backbone for Hostile Theorist and Citation Auditor.research-reportproduces thereport.mdand figure layout this skill consumes; reports that follow its conventions are easiest to review.- Korean translation of
summary.mdfollows the user's standing convention when requested: a Sonnet subagent producessummary_ko.mdnext to the English file.
Guardrails
- Never modify the draft. This skill is read-only with respect to the manuscript.
- Never silently skip a persona. If a persona cannot run (e.g. no figures to critique), say so in
summary.mdunder "skipped". - Never accept a persona output that invents findings not grounded in the draft, figures, or bibliography. If a persona returns unsupported claims, rerun that persona once with an explicit "ground every finding in a section, figure, or citation pointer" instruction; if it fails again, mark its section as unreliable in
summary.md. - Never fabricate a citation when checking citations. If
reference-searchreturns nothing for a cited title, flag it as "not found via OpenAlex — manual verification required" rather than guessing. - Never push the review outputs. Writing under
outputs/review/...is fine; git operations are out of scope.
Resources
references/personas.md— the six persona briefs with role, attack surface, ignored concerns, and output contract.references/meta-editor.md— dedup, severity ranking, and defense-experiment authoring rules.references/integration.md— wiring notes forreference-search,research-report, and Korean translation.
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.

