JohnXu22786/adversarial-review
Adversarial multi-lens code review: parallel correctness/security/maintainability lenses attack a diff, with deterministic severity grading, suppression rules and incremental review history.
gavel-review is an adversarial, multi-perspective code review plugin: independent review lenses attack a code change in parallel — each cares only about its own category (null/uninitialized, injection, complexity, ...) and maps findings to actionable checklists — then an arbitration layer merges cross-lens findings, deduplicates them, and grades severity deterministically (impact x 2 + confidence + corroboration; the same input always yields the same verdict). A deterministic model-free static sentinel runs first as a backstop, suppression rules stop the same problem class being re-reported, and a review history enables incremental comparison. An optional deep-review stage challenges each finding; refuted ones are downgraded to 'observation', never deleted. The whole process is read-only and never modifies code; zero third-party runtime dependencies.
Install
dsh plugin --profile demo add github:JohnXu22786/adversarial-reviewREADME EN body verified 2026-09-02 (repo JohnXu22786/adversarial-review, package gavel-review; zh edition link present). Install per README: dsh plugin --profile demo add github:JohnXu22786/adversarial-review (the docs use a demo profile — install into the profile where you want the gavel_review tool). Zero third-party runtime dependencies (core engine and CLI use only Node built-ins). Standalone CLI (Node >= 20.19; source runs need >= 23.6): node bin/gavel.mjs review --diff/--base/--path, --deep, --emit-rules, --out/--json-out, --fail-on.
Compatibility
DSH plugin entry registers the gavel_review tool; standalone CLI compatible with any OpenAI Chat Completions-style model service; Node >= 20.19.
Details
- Repo: JohnXu22786/adversarial-review
- Category: Development & Runtime
- Stars: 1
- Version: git github:JohnXu22786/adversarial-review
- Last push: 2026-08-17
- First seen: 2026-08-16
Recent updates
Parallel adversarial lenses; deterministic arbitration + severity; model-free static sentinel; suppression rules; review history; optional deep review; read-only; gavel_review tool + standalone CLI.
FAQ
- Does it modify my code?
- No — the whole process is read-only: lenses analyze a diff or files and the output is a report with evidence and fix suggestions.
- How are severity levels decided?
- Deterministically — impact x 2 + confidence + corroboration — so the same input always yields the same verdict; cross-lens hits on the same issue gain a corroboration bonus.
- Can I run it outside DSH?
- Yes — a standalone CLI (node bin/gavel.mjs review --diff/--base/--path) works with any model service speaking the OpenAI Chat Completions protocol.
Alternatives
PerryLink/dsh-auto-review · wuxiangru915/dsh-review-loop · CanglongCl/dsh-web-review