AgentSkillsCN

independent-review

Run N identical parallel subagents on the same task for independent validation. Use when you want consensus through redundancy — multiple agents independently analyze, review, or test the same thing, then results are compared for agreement, disagreements, and unique finds. Invocation: /independent-review [count] "task

SKILL.md
--- frontmatter
name: independent-review
description: Run N identical parallel subagents on the same task for independent validation. Use when you want consensus through redundancy — multiple agents independently analyze, review, or test the same thing, then results are compared for agreement, disagreements, and unique finds. Invocation: /independent-review [count] "task"

Independent Review

N identical agents do the same work independently. Compare results for consensus.

Current Changes

!git changes

Full Diff

!git diff HEAD

Triggers

  • •"independent review", "cross check", "validate independently"
  • •"run N agents on this", "get consensus", "replicate this analysis"
  • •Any task where independent validation adds confidence

Process

  1. •

    Parse input — Extract agent count and task from user input.

    • •/independent-review "is this migration safe?" — 3 agents (default)
    • •/independent-review 5 "review for security vulnerabilities" — 5 agents
    • •Count is always the first argument if numeric. Everything else is the task.
  2. •

    Build the prompt — Write ONE prompt. Every agent gets this exact prompt with no variation. Follow the prompt structure (Story/Business/Goal/DoD):

code
Story: {task — what the user wants analyzed/reviewed/tested and why}

Business: {constraints — codebase context, stack, what matters}

Goal: Perform this analysis independently. Be thorough. Document every
finding with evidence (file paths, line numbers, concrete examples).
Do not hedge — state your conclusions directly.

DoD:
- Every finding includes evidence (not just assertions)
- Conclusions are stated directly, not hedged
- Output is structured with clear sections
  1. •

    Dispatch N identical agents in parallel — Same subagent_type, same prompt, same tools. Use run_in_background: false so all results are collected. Name agents reviewer-1, reviewer-2, etc.

  2. •

    Synthesize — After all agents return, compare results:

    • •Consensus — Findings that 2+ agents independently identified. These are high-confidence. List each finding and which agents found it.
    • •Unique finds — Things only 1 agent caught. These need human judgment — could be an insight others missed, or a false positive.
    • •Disagreements — Where agents contradict each other. Present both sides with their evidence.
    • •Confidence — N-of-N agreement ratio (e.g., "3/3 agents agree" or "2/5 agents found this").

Key Rules

  • •Identical agents — same prompt, same tools, same model. No differentiation. Temperature and reasoning variation provide natural diversity.
  • •No use-case limits — this skill wraps any task. Code review, bug analysis, architecture assessment, test adequacy, migration risk — whatever the user provides.
  • •Evidence required — every finding must include concrete evidence. "Might be a problem" is not a finding.