doubt-driven-review
Skill0 starsChanged 5 months agoMITView on GitHub
Adversarial fresh-context review of any non-trivial technical decision or code change. MUST be used before committing any change to smart contract code (xete-tab, Xete Swap), key handling (Black Knight), encryption logic (Xete Message, MCP server), or any irreversible/immutable deployment decision. Also trigger when the user asks to "stress-test", "doubt", "red-team", or "verify" a decision. Produces a review artifact in reviews/ that the pre-commit hook checks for.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- felvieira/claude-skills-fvtheir own version, first seen Sep 2, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- felvieira/claude-skills-fv★ 28same 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 repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- gate-backtestSkill
- gate-retrospectiveSkill
- git-disciplineSkill
- solana-security-hardeningSkill
- spec-and-planSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 52
If the change under review FIXES a real defect on a protected path, also author benchmarks/BM-<name>.md (format in skills/gate-backtest) in …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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-Driven Review
Confidence is not evidence. This skill forces an adversarial second pass on decisions where being wrong is expensive or irreversible.
When this is MANDATORY (enforced by pre-commit hook)
Any staged change touching:
programs/or any Rust smart contract source- Key generation, storage, signing, or policy logic (Black Knight)
- Encryption, key exchange, or nonce handling (x25519 / AES-256-GCM paths)
- Migration or deployment scripts for immutable contracts
The loop: CLAIM → EXTRACT → DOUBT → RECONCILE
1. CLAIM
State the decision or change as a falsifiable claim. Not "improved the settlement logic" but "the commitment hash cannot be replayed across tabs because X is included in the preimage."
2. EXTRACT
List every assumption the claim rests on. For each: is it verified, inherited, or assumed? Assumptions inherited from docs or prior conversation are UNVERIFIED until checked against source.
3. DOUBT (fresh context)
Review the claim as if you had no investment in it being true. Concretely:
- Benchmarks first: check
benchmarks/BM-*.mdfor cases whosePathsoverlap this diff. Each matching benchmark's doubt prompt MUST be answered explicitly in the DDR — these are real bugs this codebase has already shipped once. Repeating a benchmarked bug because its prompt was skipped is the system's cardinal failure. - In Claude Code: spawn a subagent (or open a fresh session) with ONLY the diff, the claim, and the assumption list. No conversation history. Instruct it: "Your job is to break this. Find the input, ordering, or state that falsifies the claim."
- Cross-model escalation: for contract logic, run the same doubt prompt through the secondary model (owl-alpha) AND Claude. Disagreement between models is a finding, not noise.
- The reviewer must produce at least one concrete attack attempt per assumption, even if it fails. "Looks good" is not a review.
4. RECONCILE
For each doubt raised: refute it with evidence (test, code reference, invariant), or fix the code, or explicitly accept the risk in writing. No silent dismissals.
Required artifact
Write reviews/DDR-<short-description>-<YYYYMMDD>.md containing:
# DDR: <claim, one line>
Commit scope: <files touched>
## Claim
## Assumptions (verified / inherited / assumed)
## Doubts raised (who raised: fresh-context Claude / owl-alpha / human)
## Reconciliation (per doubt: refuted-with-evidence | fixed | risk-accepted-by-<name>)
## Verdict: SHIP | BLOCKStage this file in the same commit as the change. The pre-commit hook rejects contract-path commits without a staged reviews/DDR-*.md.
If the change under review FIXES a real defect on a protected path, also author benchmarks/BM-<name>.md (format in skills/gate-backtest) in the same commit — every real bug becomes a permanent doubt prompt so it can never ship twice unexamined.
Anti-rationalization table
| Rationalization | Reality |
| "It's a small change" | Small changes to immutable contracts are permanent small changes |
| "Tests pass" | Tests encode the same assumptions the author made |
| "We reviewed this pattern before" | The pattern was reviewed; this instance was not |
| "The other model agreed" | Agreement without an attack attempt is not review |
| "We're in a hurry for Breakpoint" | Deadline pressure is when this skill pays for itself |
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.

