Critique Sub-Agent Reviewer
Every time the main agent produces copy, research, or code, hand it to a critique sub-agent that returns fix notes before the output ships.
Before & after
Same prompt, same agent — the only difference is the skill.
Having someone review work before it goes out is important for quality. You could do it manually, but that creates bottlenecks. An automated system would mean your main agent produces output, another agent reviews it and returns feedback, and then the original agent revises. The challenge is setting up the workflow so it doesn't loop forever or miss important issues. You'd need to define what counts as a blocker versus a minor issue. How strict do you want the review criteria to be?
CRITIQUE LOOP WORKFLOW
SETUP: Two-agent handoff pattern
STEP 1: MAIN AGENT PRODUCES DRAFT
Kai (main orchestrator) receives task: "Write a product launch announcement"
Kai outputs:
• Draft copy (350 words)
• Target audience metadata
• Brand voice markers used
Kai routes to critique agent (does NOT ship yet)
STEP 2: CRITIQUE AGENT REVIEWS
Critique agent reads draft with evaluation frame:
→ Brand voice consistency: does tone match company personality?
→ Technical accuracy: no false claims?
→ Persuasion strength: compelling enough for target audience?
→ Grammar and clarity: professional standard?
→ Length: appropriate for channel?
Critique agent returns FIX NOTES (structured JSON):
[
{
"severity": "high",
"location": "paragraph 2",
"issue": "claim about performance (40% faster) not cited",
"fix": "add citation to benchmark or remove claim"
},
{
"severity": "medium",
"location": "paragraph 3",
"issue": "passive voice (we're recommending...) conflicts with brand voice",
"fix": "rewrite active: we recommend..."
},
{
"severity": "low",
"location": "final sentence",
"issue": "cliche (excited to announce) used twice in document",
"fix": "replace one with more specific language"
}
]
STEP 3: MAIN AGENT REVISES
Kai receives critique notes and edits:
• Adds benchmark citation to paragraph 2 (verifies accuracy)
• Rewrites paragraph 3 in active voice
• Replaces cliche in final sentence with specific language: "users see 40% faster results in production"
Revised draft returned to Kai for quality check against critique notes
STEP 4: APPROVAL GATE
Kai compares revised draft against critique checklist:
✓ No high-severity issues remain?
✓ Brand voice consistent?
✓ All claims cited?
✓ Length appropriate?
If ALL checks pass → SHIP
If ANY remain → Route back to critique agent for second pass (max 1 retry)
STEP 5: DELIVERY
Final copy shipped to user with metadata:
• Revision count: 1
• Critique cycles: 1
• Final QA timestamp
• Confidence score: 94%
CUSTOMIZATION OPTIONS
output_types: which artifacts get critiqued?
→ copy only? code also? research?
Critique criteria: brand voice, accuracy, brevity, tone
Max revisions: cap loops before shipping (default: 2)
Blocker severity: only revise on severe issues or fix all?
Delivery gate: only ship after critique passes 100% or 90%?
COST OPTIMIZATION
One critique pass per task = 2 API calls (draft + revision)
Average cost per task: 0.80
Monthly (100 tasks): 80
ROI: catch 5-10 errors/month that would damage brandAbout this skill
name: critique-sub-agent-reviewer description: Every time the main agent produces copy, research, or code, hand it to a critique sub-agent that returns fix notes before the output ships.
Critique Sub-Agent Reviewer
What This Does
The creator runs a Kai main-orchestrator agent that always calls a critique sub-agent to review copy, research, or code. The critique agent returns a list of fix notes, the main agent revises, and only then does the output ship, meaning the human sees near-final drafts.
What You Get
- Dedicated critique sub-agent role
- automatic hand-off after main-agent draft
- fix-notes list format
- revision loop before delivery
- works for copy, research, or code
Customize
- output_types: which artifact types get critiqued
- critique_criteria: brand voice, accuracy, brevity
- max_revisions: cap loops before shipping
- blocker_severity: only revise on severe issues
- delivery_gate: only ship after critique passes
Sample Output
The main agent writes a landing-page draft, Kai routes it to the critique sub-agent, three fix notes come back, the main agent revises, and the second version ships.
Target Audience
Agent-orchestration builders shipping high-quality output autonomously
SKILL.md preview
---
name: critique-sub-agent-reviewer
description: Use when a main agent produces copy, research, or code and needs a dedicated critique sub-agent to review it and return fix notes before it ships.
version: 1.0.0
category: Agent Orchestration
author: AgentVolt
license: proprietary
tags:
- agent-orchestration
- standard
---
# Critique Sub-Agent Reviewer
Routes every draft a main agent produces to a critique sub-agent before it ships. The critique agent returns fix notes, the main agent revises, and only the reviewed version reaches the human.
## When to use
… (sign up to view the full skill)More ai & agents skills
View all AI & Agents skills →Self-Improving Skill Overnight Loop
Run an automated overnight research loop on a Claude skill: try a tweak, run binary assertions, keep or revert based on pass rate, and iterate.
Ralph Loop Autonomous Overnight Runner
Wire the Claude Code CLI mode (-p flag) into the Ralph loop plugin so a while-loop keeps feeding the same prompt with updated files until the task is done.
Recursive Skill Failure-Fix Loop
Every time your skill fails on a real task, feed the failure back and have Claude update the skill.md so it never breaks that way again.
AI Starter Toolkit
A free, curated starting point — the essential prompts and habits to get real value from AI on day one.