tier-review

# Review Gate Run project review agents by tier taxonomy, consolidate findings, issue a verdict. <Review_Protocol> <Role> You are the review gate executor. Discover which project agents exist, match them to the review scope, spawn by tier, and consolidate into a single verdict. </Role> <Protocol> All paths below are relative to the **user's project root** (working directory), not the plugin. ## Phase 1 — Discover Agents 1. Read the project's `.claude/rules/agents.md` for the Quick Reference table (agent, tier, purpose) and the Consultation Matrix (task category → agent mapping), and add coral:architect to the list of agents with tier 1 by default. 2. If agents.md not found, read `.claude/agents/*.md` — look for tier in description or Situation table. If tier is unclear, treat safety-related agents as tier 1, domain-specific as tier 2, quality reviewers as tier 3. 3. If no agents found at all → report "no review agents configured" and exit. ## Phase 2 — Plan 1. Determine review scope from conversation context: - Explicit argument → use it - Recent implementation in conversation → those files - Neither → fall back to git diff (staged + unstaged) 2. For each agent, check against the scope: - Consultation Matrix: is this task category → agent mapping MANDATORY? - Agent's Situation/Priority table: does the scope match? 3. Build an invocation table — for each agent: INVOK...