Skip to content
Agentmart
Submit a tool

review-pass

Stage-gated multi-AI cross-validation review with optional REQ-ID traceability.

Skill by nextain

IndexedApache-2.0
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/nextain-review-pass

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Stage-gated multi-AI cross-validation review with optional REQ-ID traceability. 4 stages (planning, development, test, integration) with configurable reviewers, finding consensus, and convergence loop. Fully project-agnostic and distributable.

Use cases: testing

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Node.js package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/review-pass

What's inside

Skill name
review-pass
Description
Stage-gated multi-AI cross-validation review with optional REQ-ID traceability. 4 stages (planning, development, test, integration) with configurable reviewers, finding consensus, and convergence loop. Fully project-agnostic and distributable.
License
Apache-2.0
Instructions
87 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (29)

  • SKILL.md
  • atoms.json · examples/one-shot/
  • base.md · examples/one-shot/
  • reviewer-delta.md · examples/one-shot/
  • configuration-and-requirements.md · references/
  • consensus-and-convergence.md · references/
  • invocation-and-output.md · references/
  • preflight.md · references/
  • reporting-and-delivery.md · references/
  • stage-profiles.md · references/
  • check-output-boundary.cjs · scripts/
  • invoke-reviewer.mjs · scripts/
  • measure-complexity.mjs · scripts/
  • review-cost-log.mjs · scripts/
  • review-output-contract.mjs · scripts/
  • review-preflight.mjs · scripts/
  • validate-review-output.mjs · scripts/
  • test-context-output-lenses.cjs
  • test-distribution-parity.cjs
  • test-output-boundary.cjs
  • agy-denied-empty.ndjson · tests/fixtures/
  • agy-simple-success.ndjson · tests/fixtures/
  • agy-success-review.ndjson · tests/fixtures/
  • invoke-reviewer-cli.test.mjs · tests/
  • invoke-reviewer.test.mjs · tests/
  • measure-complexity.test.mjs · tests/
  • review-cost-log.test.mjs · tests/
  • review-output-contract.test.mjs · tests/
  • review-stream.test.mjs · tests/

SKILL.md (first lines)

# Review Pass v3 — Stage-Gated Multi-AI Cross-Validation

## Purpose

Multi-AI cross-validation review that adapts review power, perspective, and
convergence criteria to the current development stage. Optional REQ-ID
traceability for requirements-driven projects.

**Core principles:**
- **Project-agnostic**: No hardcoded paths, tools, or conventions. All configurable.
- **Independent roles**: Four evidence-separated review executions reduce shared-context blind spots; adapters and model providers are configurable
- **Stage-aware**: Planning, development, test, and integration have different needs
- **Convergent**: Automated loop until at least 2 consecutive clean rounds for every standard stage
- **Traceable**: REQ-ID coverage validated at every stage (when applicable, optional)
- **Source-complete**: In governed mode the exact source chain and full current/prior scope history are mandatory review inputs; caller-selected files cannot redefine scope
- **Baseline-preserving**: Planning and integration compare the current product with an immutable baseline and an explicit surface-preservation contract
- **Evidence-separated**: Source, baseline, implementation/test, and authority/release reviewers receive different evidence views to prevent shared anchoring
- **Safe**: Auto-fix with diff preview, rollback, and safety guard
- **Fail-closed reviewer availability**: A configured reviewer that is missing,
  unauthenticated, or unavailable fails the review pass by default. An explicit
  `--require-review false` opt-out may record `NOT_RUN` for ordinary local work,
  but that result cannot satisfy a review or release gate.

The orchestrator passes `--review-id --stage --round --reviewer-index` on every reviewer call.

## Arguments

| Argument | Required | Description |
|----------|----------|-------------|
| `stage` | **yes** | `planning` / `development` / `test` / `integration` |
| `files` | **yes** | Comma-separated file paths to review |
| `request_contract_bundle` | governed mode | Private bundle locator returned by `node scripts/request-contract.cjs review-challenge --unit <id> --writer-session <id>` |
| `source_artifacts` | planning, integration | Immutable original-source files or governed bundle references; AI summaries are not substitutes |
| `baseline_ref` | planning, integration | Immutable commit/tree for the pre-change product |
| `preservation_contract` | planning, integration | Contract containing top-level `preservation.baseline_ref/intent/surfaces/vendor_sources`; each surface carries lowercase disposition, paths, evidence IDs, and destructive authority fields |
| `incident_history` | planning, integration | Immutable chronology of the original directives, later corrections, drift evidence, and current state; a selected incident excerpt or AI summary alone is insufficient |
| `context` | recommended | What was implemented/changed, issue reference |
| `req_ids` | optional | Comma-separated REQ-IDs to validate coverage |
| `deferred_req_ids` | optional | Comma-separated REQ-IDs intentionally deferred |
| `reviewers` | optional | Override default reviewers from profile |
| `--light` | optional | 1 clean round, reduced lens set |

## When to Run

| Development Phase | Review Stage | Purpose |
|-------------------|-------------|---------|
| After Plan, before Build | `planning` | Requirements ↔ Design alignment |
| After each Build phase | `development` | Code ↔ Plan alignment + REQ coverage |
| After E2E Test | `test` | Test quality + REQ-to-test mapping |
| Before Commit/Sync | `integration` | Full E2E traceability audit |

## Core Rules

> **The orchestrator AI does NOT report intermediate results to the user.**
> **Reviewer consensus is not truth. `CONFIRMED` is only a consensus label; independently validate primary evidence.**
> **Only evidence-`ACCEPTED` findings are auto-fixed with diff preview. `REJECTED` is recorded without modification, and `UNRESOLVED` blocks CLEAN.**
> **CONTESTED findings first receive independent arbitration and source-evidence verification. Ask the user only for a remaining material decision.**
> **Only the final report is shown after convergence.**
> **Governed mode forbids `--light`, caller-only file scope, unsigned deferral, and review without the exact current request-contract bundle.**
> **Planning/integration without original source, immutable baseline, and a preservation contract is NOT CLEAN.**
> **A solo CRITICAL preservation, scope, authority, or release finding is a veto and is never auto-dismissed.**
> **An unwaived deterministic `REFACTOR_REQUIRED` result blocks CLEAN before model voting.**

## Mandatory workflow

1. Validate the input schema and detect governed mode.
2. Before **planning**, read [Preflight Gates](references/preflight.md) and [Stage Profiles and Reviewer Roles](references/stage-profiles.md) in full; apply the governed and product-preservation gates.
3. Before **development**, read those same references in full; apply the governed and deterministic-complexity gates.
4. Before **test**, read those same references in full; apply the governed and deterministic-complexity gates.
5. Before **integration**, read those same references in full; apply all governed, product-preservation, and deterministic-complexity gates.
6. Before invoking any reviewer, read [Invocation and Output Contract](references/invocation-and-output.md) and [Consensus and Convergence](references/consensus-and-convergence.md) in full.
   Use the bundled headless runner for a single-call dual prompt: byte-stable base,
   validated current-directive atom ledger, then reviewer-specific delta.
7. When resolving configuration or REQ-ID traceability, read [Configuration and Requirements](references/configuration-and-requirements.md) in full.
8. Before applying fixes, reporting a verdict, or deciding delivery eligibility, read [Reporting and Delivery](references/reporting-and-delivery.md) in full.
9. Run the stage loop until its convergence rule succeeds or a hard stop yields `NOT_CLEAN`. Never infer omitted details from this entrypoint; the linked references are normative parts of the skill.

## Reference map

- [Preflight Gates](references/preflight.md) — governed request-contract, product-preservation, and deterministic-complexity gates.
- [Stage Profiles and Reviewer Roles](references/stage-profiles.md) — reviewer counts, lenses, convergence, arbiters, and separated evidence roles.
- [Invocation and Output Contract](references/invocation-and-output.md) — encoding, prompt delivery, adapters, reviewer prompt, finding schema, and parsing.
- [Consensus and Convergence](references/consensus-and-convergence.md) — matching, vetoes, arbitration, loop algorithm, degradation, health, and anti-anchoring.
- [Configuration and Requirements](references/configuration-and-requirements.md) — config search/defaults/overrides, environment detection, REQ-ID integration, and source authority.
- [Reporting and Delivery](references/reporting-and-delivery.md) — final report schema, fix/escalation classification, light mode, and delivery gate.

From nextain/naia-adk, Apache-2.0.

More from nextain/naia-adk

Signals

Stars
0
Forks
0
Open issues
10
Last commit (this folder)
Sep 23, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from nextain

Cite this page

review-pass skill. Agent Mart. https://agentmart.dev/skills/nextain-review-pass (last updated 2026-10-24)