Hermes Agent /review: Your Built-In Reviewer

The Hermes Agent /review command gives you a second pair of eyes on demand: /review [instructions] spawns an independent, full-privilege reviewer subagent that digs into your recent work — PRs, code, docs, artefacts — and delivers a full review back into your session. Here’s exactly what the docs promise, and how to make the reviewer earn its keep.

What the Hermes Agent /review command does

The documented behaviour is unusually muscular: the reviewer is an independent subagent with full privileges. It doesn’t skim your chat and offer opinions — it opens the PR, reads the actual diff, and runs code. When it’s done, the complete review re-enters your session as a background-subagent completion, so you keep working while it investigates.

The optional instructions argument steers focus: /review check the error handling paths gets you a specialist pass instead of a general one. And via auxiliary.review in config you can pin which model does the reviewing — meaning your reviewer can be a different, even stronger model than the one that wrote the work.

Hermes Agent /review in a real pipeline

The independence is the point. A model reviewing its own fresh output inherits its own assumptions; a separately-spawned reviewer — especially a different pinned model — breaks that correlation. It’s the closest thing to a colleague’s code review you can summon with one line.

Where /review fits the command family

It completes the arc that /plan starts: plan first, execute against the plan (ideally as a /goal with checkable criteria), then /review the result against intent. Because the reviewer runs in the background, the cost of the habit is nearly zero — fire it, keep moving, read the verdict when it lands.

Practical habit worth building: review before you consider work finished, not after something breaks. The one-line command removes every excuse the busy version of you usually has.

The bottom line on Hermes Agent /review

Hermes Agent /review is a genuine reviewer, not a summariser: independent, full-privilege, diff-reading, code-running, and pinnable to a model of your choosing. Make it the default final step on anything that ships and you’ve institutionalised the discipline most solo operators never get — a second opinion that always shows up.

Acting on a review like a lead, not a rubber stamp

The reviewer’s report is where the value is won or lost. Triage it in three passes: anything correctness-critical gets fixed now; style and structure notes get batched; and anything you disagree with gets a decision, not a shrug — either act on it or note why not, because silently ignoring findings trains you to stop reading them. On significant work, close the loop: fix, then /review again with focus on the changed area.

Resist the urge to argue with the reviewer inside your head. It read the actual diff and ran the actual code — on those facts it’s usually right; on taste, you’re allowed to overrule it. Keeping that line clear is what makes an always-available reviewer a discipline rather than a nag.

Last practical note: reviews compound when you feed them forward. When a reviewer catches the same class of issue twice — missing error handling, weak headings, unverified links — that’s no longer feedback about the work; it’s feedback about the process. Fold it into your prompts, your skills or your plan templates so the third occurrence never happens. A reviewer that only fixes artefacts is useful; one whose findings upgrade the system that produces the artefacts is how solo operators end up shipping at team quality.

FAQ: hermes agent /review

What does /review do in Hermes Agent?

It spawns an independent, full-privilege reviewer subagent that investigates recent work — opening PRs, reading diffs, running code — and returns a full review into your session, per the docs.

Can I direct what the reviewer focuses on?

Yes — /review takes optional instructions to steer the review’s focus.

Can a different model do the reviewing?

Yes — pin the reviewer model via auxiliary.review in config.

Does /review block my session while it works?

No — it investigates in the background and its review re-enters as a background-subagent completion.

What can it review?

Recent work broadly: PRs, code, docs and artefacts, per the documentation.

Where does /review work?

CLI and messaging gateway; on Slack use /hermes review.

Related reading

• Hermes Agent /plan: Plans Without Execution
• Hermes Agent MoA: Mixture of Agents, Explained
• Hermes Agent /refine: Self-Improvement On Demand

Last updated September 2026. This is the living guide to hermes agent /review — it gets updated as the tools change.