Skip to content
Get packs

Guide · Agent skill

How to use the Review Team Pod Reviewer agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The review-team skill is an agent skill from the OpenRig project that teaches an AI agent to act as a fresh-eyes reviewer on a software change inside a multi-agent team. It is written for developers who run OpenRig pods and want a dedicated review seat that reports findings in a consistent, severity-rated format. By the end of this guide you will know what the skill changes about the agent's behaviour, how to install it, how to trigger it, and what it needs before it is useful.

Review Team Pod Reviewermvschwarz/openrig · Apache-2.0

Quick start

  1. 1Download the ZIP. Download the review-team skill ZIP from this page.
  2. 2Install the skill. Unzip it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  3. 3Confirm your rig environment. Run rig whoami --json to confirm the rig CLI works and that the agent can identify its seat.
  4. 4Make an explicit assignment. Tell the agent which change or wave it is reviewing and who the owner is, so the skill treats the request as a real assignment.
  5. 5Let the agent read context first. The skill has the agent read CLAUDE.md, architecture docs and relevant specs before forming a position.
  6. 6Collect the review file. Check docs/review/<review-name>/ for the agent's findings file.
  7. 7Act on the severity-rated findings. Work through MUST-FIX and HIGH items first, and route anything unrelated to a separate task.

What is an agent skill, and what does review-team make the agent do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific kind of task. When the skill is installed and relevant, the agent reads these instructions and follows them. The review-team skill is one such file, named review-team, with the description: use when assigned a review or when an authored review boundary is reached.

Once active, the skill tells the agent to behave as a reviewer rather than an implementer. It covers when a review is actually assigned, how much rigor to apply to a given change, what context to read before forming a view, and how to check for duplicated or inconsistent code. It also describes writing findings to disk and reporting them to an orchestrator with rig commands, using severity ratings from MUST-FIX down to INFO.

  • Starts a review only on an explicit owner assignment or an authored component or wave review event
  • Matches depth to consequence: a small diff gets a focused pass, and a deep review method runs only when explicitly selected
  • Checks for duplication and pattern divergence, asking whether an agent in three months would find two ways to do the same thing
  • Writes artifacts to docs/review/<review-name>/ and reports findings via rig send or rig chatroom send

Install the review-team skill

You can install the skill in two ways. For Claude Code, download the ZIP of the skill from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder to share it with one repository.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into the project's .claude/skills/
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what the skill needs before you start

The skill assumes an OpenRig environment. It requires OpenRig with the rig CLI installed and a running rig environment, plus access to the project's docs, such as CLAUDE.md and the architecture and spec docs the skill references.

The skill's entry procedure runs rig whoami --json and then resolves the project.yaml, mission.yaml, active slice.yaml and selected addresses. If your repository has no pod structure, the skill has little to work with.

  • OpenRig (rig CLI) installed and a running rig environment
  • Access to CLAUDE.md, architecture docs and spec docs
  • The skill runs scripts and does not use the network, according to its listing
rig whoami --json

Trigger the skill with natural requests

The skill is meant to activate when the agent is assigned a review or reaches an authored review boundary. Ask for the review in terms of the assignment rather than a general code check, because the skill is written to refuse reviews that nobody assigned.

The skill also says that a visible milestone, an idle queue or an available reviewer is not an assignment. Name the target and the owner so the agent can tell whether the review is selected.

  • "You are the reviewer for the wave 2 change. Review the working tree against the spec and report findings to the orchestrator."
  • "Do a hindsight review of the dogfood fixes in commits since the last tag. There is no spec, so reconstruct intent from commit messages."
  • "Review the authored component boundary for the rig send changes and write your findings to docs/review."

Read the report format

Findings are written to a review file and sent to the orchestrator in a structured line format. Each finding carries a severity, a file and line, the user consequence, how the reviewer knows, and whether the change introduces, exposes or leaves the issue unchanged.

Severity levels run from MUST-FIX for broken behaviour or a real exposure on a reachable path, through HIGH and MEDIUM, to LOW and INFO, which the skill says is not a defect.

  • MUST-FIX: broken behaviour or real exposure on a reachable path
  • HIGH: contract violation or honesty failure
  • MEDIUM: real concern affecting maintenance or agent UX
  • LOW: polish, robustness or minor inconsistency
  • INFO: observation worth noting, not a defect
rig send <orchestrator-session> "REVIEW: <title> HIGH :: <file:line> :: <user consequence> :: <how you know> :: introduces|exposes|leaves" --verify

Safety notes: scripts and shell access

The listing says this skill runs scripts. The skill expects validation runs and rig CLI calls, so do not use it in a repository where the review must not run shell commands. The skill also says not to run code where the assignment forbids it, and to say so when uncertainty remains because of that restriction.

The skill does not use the network, according to its listing. It also tells the agent not to start reviews on its own, which limits unexpected activity, but you should still review the commands the agent runs in your environment.

  • Scripts: yes. Expect shell and rig CLI calls during reviews.
  • Network: no, according to the listing.
  • Restrict the agent's shell access if your repository forbids running code during review.

What people use it for

  • Giving an agent a dedicated reviewer seat in an OpenRig pod for a wave of implementation work
  • Running a hindsight review on ad hoc or dogfood fixes that were never written up as a spec
  • Getting consistent severity-rated review reports from agents across several changes
  • Checking new code for duplication against existing patterns in a multi-agent codebase

Frequently asked questions

Is the review-team skill free to use?

The source gives the licence as Apache-2.0. It does not list prices, so check the licence file in the repository for the terms that apply to your use.

What do I need to install review-team?

You need the SKILL.md skill folder, which you download as a ZIP from this page. You also need OpenRig with the rig CLI and a running rig environment, because the skill's instructions depend on rig commands.

Can I use review-team without OpenRig?

The skill is built around rig commands and a pod structure. The source does not describe a standalone mode, so for a quick review with no team or orchestrator, another review approach is likely a better fit.

Where do I install the skill in Claude?

For Claude Code, unzip the skill into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

Why does the agent say it is not starting a review?

The skill tells agents to start a review only on an explicit owner assignment or an authored review event. A milestone or an idle reviewer does not count, so name the target and the owner in your request.

Does the skill run code or use the network?

The listing says it runs scripts and does not use the network. It expects validation runs and rig CLI calls, so it is unsuitable where the review must not run shell commands.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use review-team if you run OpenRig multi-agent pods and want a dedicated reviewer seat that produces consistent, severity-rated reports and writes them to disk. Pick something else if you want a quick one-off code review with no team, no orchestrator, or no shell access, because the skill assumes all three.

Download the skill