How do I install the Review Spec skill?

Run npx skillmds@latest add tomevault-io/review-spec in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Review Spec safe to use?

This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Review Spec?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Review Spec free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Review Spec?

tomevault-io (@tomevault-io) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Review Spec by tomevault-io — Agent Skill

Name: Review Spec
Author: tomevault-io

Review Spec

Multi-model cross-review for spec documents. Uses fresh Claude context + Gemini to independently review requirements.md, design.md, and tasks.md, then merges findings into a unified report with consensus markers. Use when this capability is needed.

tomevault-io ac2d156 2 files · 3.9 KB Updated Sep 17, 2026

File contents

Review Spec — Multi-Model Cross-Review

Review a spec directory (requirements.md + design.md + tasks.md) using multiple AI models to catch issues that a single model would miss.

Why Multi-Model Review?

The same model that wrote the spec will tend to confirm its own work. Different models have different blind spots — cross-review catches structural issues, logical contradictions, and missing edge cases that single-model review misses.

Instructions

Step 1: Locate the Spec

If $ARGUMENTS is provided, use it as the spec directory path. Otherwise, search for spec directories:

find . -path "*specs*" -name "requirements.md" -o -path "*specs*" -name "design.md" | head -20

Ask the user to confirm which spec to review.

Step 2: Read All Spec Files

Read the following files from the spec directory:

requirements.md (if exists)
design.md (if exists)
tasks.md (if exists)

If any file is missing, note it as a finding.

Step 3: Self-Clarify (Quick Internal Check)

Before calling external models, do a quick consistency check:

Does every requirement have a corresponding design section?
Does every design component have corresponding tasks?
Are task dependencies ordered correctly (no circular deps)?
Are [HUMAN] tags used appropriately?

List any contradictions found.

Step 4: Run Cross-Review Script

Execute the review script:

bash ${CLAUDE_SKILL_DIR}/../scripts/review-spec.sh "<spec-directory>"

This script:

Sends specs to a fresh Claude instance (zero shared context)
Sends specs to Gemini 3 Pro (different model, different blind spots)
Merges both reviews into a final report

Step 5: Present Results

Read the generated reviews/review-final.md and present to the user:

Show Critical issues first (both reviewers agree)
Show Major issues (one reviewer flagged)
Show Minor suggestions
Highlight items marked as Consensus (highest confidence)

Step 6: Offer Next Steps

Ask the user:

Fix the Critical issues now?
Re-run review after fixes?
Proceed to pipeline execution?

Review Checklist (Built into the Review Prompt)

Spec Consistency

Every requirement has a corresponding design module
Every design module has corresponding tasks
[HUMAN] tags are justified (not marking AI-doable tasks as HUMAN)

Technical Risk

External dependencies are specified (API versions, SDK versions)
Performance risks identified (large data, concurrency, timeouts)
Security risks addressed (permissions, data leaks, injection)

Executability

Every task has a testable acceptance criterion
Task granularity fits single pipeline module execution
Task dependency order is correct

Completeness

Error handling and edge cases covered
Test tasks exist for each functional task
Data migration / backward compatibility (if applicable)

Source: yinshucheng/claude-skills — distributed by TomeVault.

---name: review-specdescription: Multi-model cross-review for spec documents. Uses fresh Claude context + Gemini to independently review requirements.md, design.md, and tasks.md, then merges findings into a unified report with consensus markers. Use when this capability is needed.---# Review Spec — Multi-Model Cross-ReviewReview a spec directory (requirements.md + design.md + tasks.md) using multiple AI models to catch issues that a single model would miss.## Why Multi-Model Review?The same model that wrote the spec will tend to confirm its own work. Different models have different blind spots — cross-review catches structural issues, logical contradictions, and missing edge cases that single-model review misses.## Instructions### Step 1: Locate the SpecIf `$ARGUMENTS` is provided, use it as the spec directory path.Otherwise, search for spec directories:```bashfind . -path "*specs*" -name "requirements.md" -o -path "*specs*" -name "design.md" | head -20```Ask the user to confirm which spec to review.### Step 2: Read All Spec FilesRead the following files from the spec directory:- `requirements.md` (if exists)- `design.md` (if exists)- `tasks.md` (if exists)If any file is missing, note it as a finding.### Step 3: Self-Clarify (Quick Internal Check)Before calling external models, do a quick consistency check:- Does every requirement have a corresponding design section?- Does every design component have corresponding tasks?- Are task dependencies ordered correctly (no circular deps)?- Are `[HUMAN]` tags used appropriately?List any contradictions found.### Step 4: Run Cross-Review ScriptExecute the review script:```bashbash ${CLAUDE_SKILL_DIR}/../scripts/review-spec.sh "<spec-directory>"```This script:1. Sends specs to a fresh Claude instance (zero shared context)2. Sends specs to Gemini 3 Pro (different model, different blind spots)3. Merges both reviews into a final report### Step 5: Present ResultsRead the generated `reviews/review-final.md` and present to the user:1. Show **Critical** issues first (both reviewers agree)2. Show **Major** issues (one reviewer flagged)3. Show **Minor** suggestions4. Highlight items marked as **Consensus** (highest confidence)### Step 6: Offer Next StepsAsk the user:- Fix the Critical issues now?- Re-run review after fixes?- Proceed to pipeline execution?## Review Checklist (Built into the Review Prompt)### Spec Consistency- Every requirement has a corresponding design module- Every design module has corresponding tasks- `[HUMAN]` tags are justified (not marking AI-doable tasks as HUMAN)### Technical Risk- External dependencies are specified (API versions, SDK versions)- Performance risks identified (large data, concurrency, timeouts)- Security risks addressed (permissions, data leaks, injection)### Executability- Every task has a testable acceptance criterion- Task granularity fits single pipeline module execution- Task dependency order is correct### Completeness- Error handling and edge cases covered- Test tasks exist for each functional task- Data migration / backward compatibility (if applicable)---> Source: [yinshucheng/claude-skills](https://github.com/yinshucheng/claude-skills) — distributed by [TomeVault](https://tomevault.io).

Frequently asked questions