Chorus v0.6.1: Your Time Costs More Than Tokens. /yolo it!
Chorus v0.6.1 is here! This release does one thing: break you out of review-every-Claude-Code-plan hell.
Review Hell
If you’ve ever used Claude Code to build a full feature, you know the loop. Hand it a task, wait for it to finish, review the plan, spot something wrong, send it back, review again, confirm it’s fine, hand it the next task. Repeat. Repeat. Repeat.
The agent writes code. You review full-time.
Tokens are cheap, but the real bottleneck is you. Every plan needs your eyes. Every output needs your sign-off. One moment of inattention and a flawed design slips through. No matter how many agents you run in parallel, they’re all stuck behind your single-threaded review bandwidth.
This isn’t a model problem. The model is smart enough. It’s a harness problem: there’s no quality gate between the agent and you, so you have no choice but to watch every step yourself.
Today Chorus fills that gap, so you can /yolo.
/yolo: One Prompt, Start to Finish
v0.6.1 adds a new command to the Chorus Claude Code plugin:
/yolo Add dark mode to the project with automatic system preference detectionOne sentence in, the agent runs the entire pipeline automatically: Idea → Elaboration → Proposal → Review → Task breakdown → Parallel execution → Verification → Done. It won’t ask you a single question along the way.
Sounds reckless, but /yolo isn’t flying blind. It can go full-auto because the Chorus workflow has three layers of guardrails backing it up.
Layer 1: Elaboration — Surface the Assumptions
After receiving your one-liner, the agent doesn’t jump straight into writing a plan. It enters an Elaboration phase first, generating a set of key questions: Which browsers need support? Should data be persisted? What are the performance targets?
In the normal workflow, a human answers these. In /yolo mode the agent answers them itself, reasoning from the original prompt to pick the most sensible options.
The value here isn’t the accuracy of the answers. It’s that the agent’s assumptions are now explicit and on the record. Every self-answered question is logged in the Chorus audit trail. What did the agent assume? Why did it pick A over B? All traceable. If the final delivery is off, you trace back to Elaboration and pinpoint where the assumption diverged, instead of staring at code trying to guess what the agent was thinking.
Full automation doesn’t mean black box.
Layer 2: Proposal — Think Before You Code
After Elaboration, the agent still doesn’t start writing code. It produces a full Proposal: tech design documents, task breakdown, dependency DAG, and acceptance criteria for every task.
The proposal isn’t free-form text. It’s structured: Documents carry the technical design, Tasks define units of work, Acceptance Criteria define what “done” looks like. Once approved, these turn directly into executable work items. The agent doesn’t need to re-interpret its own plan.
When nobody is in the loop to course-correct in real time, the quality of the plan determines the quality of everything downstream. A bad Proposal plus five agents executing in parallel just amplifies the mistake five times over. That’s why /yolo is actually most careful at this step: get the planning right, then let it rip.
Layer 3: Reviewer — Let Agents Review Agents
The Proposal doesn’t get rubber-stamped. A proposal-reviewer kicks in automatically. It checks whether API contracts are consistent with upstream and downstream consumers, whether task granularity makes sense, whether acceptance criteria are actually testable. Problems? Rejected with specific BLOCKERs. The agent fixes them and resubmits. The reviewer reviews again. Up to three rounds. If it still doesn’t pass, it escalates to a human.
Same thing after task execution. A task-reviewer checks whether the code actually meets acceptance criteria. Doesn’t pass? Sent back for another round. Same three-round cap.
All that review work you used to do yourself? An agent handles the bulk of it now. Interface mismatches, unmet criteria, dependency cycles — these mechanical checks no longer cost your attention. By the time something reaches you, it’s already survived a round of adversarial review. You just make the final call.
Review isn’t gone. It’s just not your job anymore.
The Industry Is Converging
Last week Anthropic shipped two things worth reading side by side.
Advisor Tool lets Sonnet consult Opus mid-execution. One does the work, the other checks the work. Essentially a higher-tier reviewer for agents. This solves the same class of problem as Chorus’s Review Agent, except Advisor operates at the model level (within a single API call) while Chorus’s Reviewer operates at the workflow level (separate process, separate context, review results permanently archived).
UltraPlan has Claude Code generate a full implementation plan in the cloud. Humans comment on sections, request revisions, iterate until satisfied, then execute. This solves the same problem as Chorus’s Proposal mechanism, except Chorus’s Proposals are structured (Document + Task DAG + AC) and turn directly into executable work items upon approval.
One is “someone watching while you execute.” The other is “think it through before you start.” The industry’s view on agent workflows is converging: execute fast, but plan carefully; automate deep, but never skip review.
So, /yolo
Elaboration surfaces the assumptions. Proposal locks down the plan. Reviewer guards the quality. Stack all three and full automation stops being reckless. It becomes something you can actually trust.
/yolo isn’t “we finally built full automation.” It’s “we finally have the guardrails to pull it off.” Your time shouldn’t be spent reviewing agent output line by line.
Ctrl+C to bail out any time. Everything created so far stays in Chorus. Pick up with /develop or /review to take over manually. If a task fails review repeatedly (three-round cap), the pipeline doesn’t stall. It flags the task for human intervention and keeps going.
Try It
/yolo whatever feature you wantv0.6.1 is on GitHub Releases. /yolo ships as a Chorus Plugin skill. Make sure to upgrade the plugin to v0.7.0. Your API key needs all three roles: pm_agent, admin_agent, developer_agent.
Questions or feedback? GitHub Issues or Discussions.
GitHub: Chorus-AIDLC/Chorus | Release: v0.6.1