Installs into .claude/skills of the current project.
Are you the author of Ztu Agent Arbitration?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/realjustawall-ztu-agent-arbitration)
---
name: ztu-agent-arbitration
description: "Merge independent critic advocate brand a11y 3D and performance findings, require verifiable evidence and resolve contradictory actions."
---
# Multi-Agent Evidence Arbitration 2.0 — Perfect_AI VERIFIED ULTIMATE
## Mission / ماموریت
Merge independent critic advocate brand a11y 3D and performance findings, require verifiable evidence and resolve contradictory actions.
## When to invoke
After parallel multi-agent design reviews and before code changes.
## Exact workflow
1. Run role specialists read-only and collect JSON reports.
2. Validate evidence paths, confidence, affected component and suggested action.
3. Group repeated findings and corroborate cross-role agreement.
4. Identify explicit conflicts: e.g. add motion versus reduce motion on same component.
5. Arbitrator issues accept/reject/investigate with evidence and brand priority, not majority vote only.
6. Produce approved action queue; no autonomous edits without explicit authorization.
## Acceptance criteria
- Duplicate findings become one tracked issue.
- Conflicts are not silently resolved.
- Missing evidence never passes critical gates.
- Reports trace reviewer identity and source files.
## Evidence contract
Write a machine-readable report with `status` = `pass|fail|unverified`, measured values, runtime/browser/device, changed files, exact command, artifacts and rationale. **Never** claim hardware/browser success from unit tests alone. Preserve original project files unless user explicitly authorizes edits. Avoid MCP; use local CLI and files.
## Compatibility and precautions
Read the installed dependency versions and project conventions first; prefer official APIs. Respect reduced motion, keyboard navigation, privacy, Persian RTL and low-power devices. Do not let two animation engines own the same DOM property. Test fallback modes and undo changes when the same acceptance check regresses.
## Working files
- `../../references/agent-arbitration-implementation.md`
- `../../examples/agent-arbitration/README.md`
## Official references
- https://playwright.dev/docs/test-projects