reversa-reviewer

Review reverse-engineered specifications and reconcile cross-unit contradictions.

1.5k|383|Updated Apr 26, 2026
One-click install
npx skills add https://github.com/sandeco/reversa --skill reversa-reviewer-sandeco
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: reversa-reviewer
Source: https://github.com/sandeco/reversa/tree/main/agents/reversa-reviewer
Command: npx skills add https://github.com/sandeco/reversa --skill reversa-reviewer-sandeco

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill prevents AI-generated reverse-engineering specifications from being wrong, incomplete, or internally inconsistent by enforcing a structured review workflow with confidence scoring and validation questions.

Core Features & Use Cases

  • Cross-unit and per-unit consistency review: checks canonical files in each unit (requirements/design/tasks) plus optional artifacts for gaps and contradictions.
  • Confidence reclassification: upgrades/downgrades statement certainty (🟢 CONFIRMADO, 🟡 INFERIDO, 🔴 LACUNA) using defined confidence rules.
  • Human validation loop: generates _reversa_sdd/questions.md and processes user answers to update specs and produce a final confidence report.
  • Optional Codex cross-review: when available and the doc level requires/permits it, delegates an independent review and merges results.

Quick Start

Tell the Reviewer to start the revision phase by pointing it to your Reversa output folder so it can read state/config and review each generated unit spec.

Frequently Asked Questions about reversa-reviewer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I validate AI-generated specifications for legacy modernization?

Legacy specification validation involves reviewing per-unit canonical files and optional artifacts to reconcile cross-unit contradictions and matrix completeness. It applies confidence rules to reclassify statements and generates validation questions for identified gaps.

How does confidence scoring work for reverse-engineered requirements?

Confidence scoring for reverse-engineered requirements works by upgrading or downgrading statement certainty into categories like confirmed, inferred, or gap. It uses defined confidence rules during the revision phase to assess the reliability of each generated specification statement.

What is the best way to review reverse-engineered specifications for cross-unit contradictions?

The best way to review reverse-engineered specifications for cross-unit contradictions is to read the Reversa state and config, then review each generated unit spec. This process reconciles inconsistencies across requirements, design, and tasks canonical files plus optional artifacts.

Can I use Codex cross-review for legacy specification validation?

Yes, you can use Codex cross-review for legacy specification validation when the documentation level requires or permits it. The process delegates an independent review of the reverse-engineered specs and then merges the results back into the final output.

Do I need a specific folder structure to start validating legacy-to-spec outputs?

Yes, you need to point the reviewer to your Reversa output folder so it can read state and config files. This allows it to access the per-unit canonical files and generate the final confidence report in the designated internal directory.

Why does my reverse-engineered specification have incomplete requirements?

Reverse-engineered specifications often have incomplete requirements due to internal inconsistencies or missing information from the legacy system. A structured validation workflow identifies these gaps using confidence scoring and generates a human validation loop to resolve them.