Skip to main content
Coding agents (Claude Code, Cursor, Windsurf, etc.) and autonomous agents (OpenClaw, Hermes) can manage the review process for you. When a curation triggers a review, the agent sees the pending changes and can approve or reject them — with your permission.

​
How agents discover reviews

There are two paths depending on whether the agent waits for curation to complete.

​
Inline notification (default)

When an agent runs brv curate without the --detach flag, it waits for the curation to finish. If any operations are flagged for review, the review summary is printed directly in the command output:
The output includes:
  • The task ID
  • Each flagged operation with its type, impact, reason, and before/after summaries
  • The exact brv review approve and brv review reject commands
The agent reads this output, presents the changes to you, and asks for your decision.

​
Detached mode (--detach)

When an agent runs brv curate --detach, the command queues the curation and returns immediately without waiting for completion:
Since the agent doesn’t wait, it never sees the inline review notification. Instead, it uses brv review pending to check for unresolved reviews after the task completes:
This is the standard pattern for MCP-based integrations, where the coding agent calls brv-curate (which uses detached mode internally) and then follows up with brv review pending.

​
The agent review loop

A typical agent-driven review follows this sequence:

​
Agent review in practice

Any agent with a ByteRover connector (skill, MCP, rules, or hook) follows the same pattern. The connector teaches the agent the brv review commands, so it can:
  • List pending reviews with brv review pending
  • Approve or reject with brv review approve/reject <taskId>
  • Selectively approve/reject files with --file
  • Use --format json for structured parsing
The agent reads the review summary, presents the changes to you conversationally, and asks for your decision before running the approve or reject command. Example — OpenClaw handling a review:
1

User asks OpenClaw to curate

2

OpenClaw curates and discovers a pending review

OpenClaw

I’ve curated the caching strategy update, but ByteRover flagged it as a high-impact change that needs your review.This changes your project’s caching approach from Redis to Memcached. Should I approve or reject this change?
3

User responds

4

OpenClaw approves the change

OpenClaw

Done — approved infrastructure/caching/caching_strategy.md. The change is now finalized and eligible for push.

​
Detach mode reference

The --detach flag is useful when you don’t want the agent to block while curation processes. The tradeoff is that the agent must explicitly check for reviews afterward using brv review pending.
If your agent runs unattended and you don’t want to review changes at all, you can disable the HITL review log for the current project with brv review --disable. See Disable the review log for the full behavior.

​
Connector setup

To enable agent-driven reviews, your agent needs a ByteRover connector installed. The connector teaches the agent the brv review commands alongside brv query and brv curate.
For setup instructions, see Coding Agent Connectors and Autonomous Agents.