multi-agent-review: a Claude Code skill by riekelt
For teams who want a written spec or plan checked by several reviewers before any code is built. Skip it if you want code reviewed, since it only reads specs and plans.
High-risk warning At least one scanner rated it high risk, so we keep it out of the top ten. Read its code before you install it.
- Score
- 59of 100
- Rank
- #225of 239
- Installs
- 23.6Kon skills.sh
- Security
- Warning 3 scanners
- Updated
- 2026
- Context
- ~137tokens always
Install
npx skills add https://github.com/riekelt/multi-agent-review --skill multi-agent-review Run it in your terminal. View the source
Using the Claude app rather than Claude Code? You can add a skill there too. How to add it in the app
What does multi-agent-review do?
multi-agent-review is a skill that runs a panel of six AI reviewers over a spec or plan before work starts. You call it with spec or plan mode; it finds the newest file in docs/superpowers/ or the path you give, checks it exists, and reads project rules from project-rules.md or CLAUDE.md. Two model tiers (Haiku and Sonnet by default) each review completeness, alignment and risk in parallel. Where they disagree, an Opus juror rules. The result is a Blockers, Warnings or Observations verdict plus a JSON block. Blockers stop the next step unless you override, which is logged as a git note. Clean results hand off to the next skill.
When should you use multi-agent-review?
Use it when
- A spec is written and you want it checked before planning
- A plan needs a second opinion before agents execute it
- You want a recorded gate before risky work starts
Skip it when
- You need a review of code, not documents
- You don't keep specs or plans as Markdown files
- You want to avoid the cost of seven agent runs
What it needs and what it can touch
You need
- Specs or plans as Markdown, ideally under docs/superpowers/
- The writing-plans and subagent-driven-development skills for the handoff
What it can do
- Runs shell commands (ls, jq, git notes)
- Reads project files, including CLAUDE.md and project-rules.md
- Starts up to seven subagents on different models
- Writes a git note when you override blockers
From reading its instructions, not from running it.
Why it scores 59
Strongest on quality (100 of 100), weakest on adoption (37).
Is multi-agent-review safe to install?
At least one scanner rated it high risk, so we keep it out of the top ten. Read its code before you install it. Scanners often miss what other scanners catch, and a skill can change after a scan. Read its source before you install it.
| Scanner | Result | Date |
|---|---|---|
| Gen Agent Trust Hub | Passed | Aug 23, 2026 |
| Socket | Passed | Aug 23, 2026 |
| Snyk | High risk | Aug 23, 2026 |
Details
- Installs
- 23,606
- Repository stars
- 60
- Last repository change
- Tokens always loaded
- about 137
- Tokens when used
- about 4,296
- License
- Not stated
- Data confidence
- High
Did it do the job for you?
Alternatives to multi-agent-review
All skills for this job-
Get Elephant Goldfish ModelElephant Goldfish Model
For engineers who write design docs before building and want fresh-eyed agent reviewers.
-
Get spec-driven-implementationspec-driven-implementation
For teams building large features who want PRODUCT.md and TECH.md specs checked in before and during coding.
- Get write-product-spec
write-product-spec
For teams that want a behavior-first PRODUCT.md spec written before a feature is built.
-
Get brainstormingbrainstorming
For anyone who wants Claude to pin down what they really want before it builds.
multi-agent-review Get multi-agent-review