doubt-review
Skill0 starsChanged 5 months agoMITView on GitHub
Adversarial fresh-context review for high-stakes irreversible decisions
In this repository26 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/000-project.mdcCursor rule
- .cursor/rules/accessibility.mdcCursor rule
- .cursor/rules/api-integration.mdcCursor rule
- .cursor/rules/architecture.mdcCursor rule
- .cursor/rules/dart-style.mdcCursor rule
- .cursor/rules/design-system.mdcCursor rule
- .cursor/rules/flutter-ui.mdcCursor rule
- .cursor/rules/security.mdcCursor rule
- .cursor/rules/state-management.mdcCursor rule
- .cursor/rules/testing.mdcCursor rule
- fix-dependenciesSkill
- flutter-componentSkill
- flutter-implementSkill
- generate-mocksSkill
- grind-until-greenSkill
- implement-tddSkill
- monitor-terminalSkill
- plan-featureSkill
- repo-navigatorSkill
- review-codeSkill
- save-contextSkill
- suggest-ruleSkill
- write-commitSkill
- write-readmeSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 52
Do not pass your CLAIM or reasoning — that biases agreement.
- Distinctive: line 54
If subagents are unavailable: ask the user to review, or run degraded self-questioning (flag as degraded — not fresh context).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Doubt review
Challenge non-trivial decisions before they ship. Complements review-code (post-hoc PR review).
See AGENTS.md — normal work is one agent; this skill allows one separate reviewer (fresh context), not a simulated team.
When to use
- Auth, payments, migrations, data deletion
- Architectural or security decisions under uncertainty
- Irreversible blast radius (prod deploy, public API, RLS changes)
When NOT to use
- Formatting, renames, obvious one-liners
- User asked for speed over verification
Process
1. CLAIM — state the decision + why it matters (2–3 lines)
2. EXTRACT — smallest artifact (diff, function, proposal) + contract (what must hold)
3. DOUBT — fresh-context adversarial review (see below)
4. RECONCILE — classify each finding: contract misread / actionable / trade-off / noise
5. STOP — trivial findings only, 3 cycles max, or user says shipStep 3 — Cursor implementation
Spawn one Task subagent (readonly: true) with this prompt only — no session history, no CLAIM:
Adversarial review. Find what is wrong with this artifact.
Assume the author is overconfident. Look for:
- Unstated assumptions
- Edge cases not handled
- Hidden coupling or shared state
- Failure modes under unexpected input
- Violations of repo rules (AGENTS.md, architecture.mdc, security.mdc)
Do NOT validate. Find issues, or state explicitly that none found after thorough examination.
ARTIFACT: <paste>
CONTRACT: <paste constraints>Do not pass your CLAIM or reasoning — that biases agreement.
If subagents are unavailable: ask the user to review, or run degraded self-questioning (flag as degraded — not fresh context).
Reconcile
You remain the implementer. Classify findings; fix actionable items; document accepted trade-offs.
Red flags
- Skipping doubt on migrations/auth then reviewing only at PR time
- Passing CLAIM to the reviewer
- More than 3 cycles without escalating to user
- Using doubt for trivial changes
Related
review-code— finished artifact checklistimplement-tdd— failing test satisfies doubt for behavioral claims.cursor/rules/security.mdc— auth, tokens, RLS
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.

