Agents on Steroids
Expanded from 8 to 15 agents covering the complete sprint cycle: Think, Plan, Build, Review, Test, Ship, Reflect. Zero dependencies.
7 new agents
- Strategist — product strategy with anti-sycophancy and forcing questions
- Spec Writer — vague intent to precise 7-section specs with acceptance criteria
- Investigator — root-cause debugging, no fixes without diagnosis
- Reviewer — code review with 7-specialist parallel review army
- Shipper — tests, changelog, version bump, PR in one command
- Retro — sprint retrospectives with persistent learnings
- Context Manager — session persistence, pause and resume any sprint
5 enhanced existing agents
- Architect: +ENG-REVIEW mode (architecture lock-in checkpoint)
- UX: +AUDIT mode (8-dimension design scoring 0-10)
- DevOps: +CANARY mode (post-deploy monitoring)
- QA: +BROWSER and +REPORT modes
- Security: +AUDIT mode (OWASP Top 10 + STRIDE)
3 sprint modes
/agency-run FULL <task>— full cycle with all 15 agents/agency-run QUICK <task>— fast track: plan, build, test, ship/agency-run HOTFIX <task>— investigate, fix, ship
New infrastructure
- ETHOS.md — 3 agnostic builder principles referenced by every agent
- Eval system — agent quality scoring (14/14 pass) + 7 runtime eval tasks
- Multi-platform installer — Claude Code, Cursor, Copilot CLI, Codex
- ARCHITECTURE.md — system design overview
Upgrade
Plugin users: /reload-plugins to get v2.0.0 automatically.
Manual users: git pull && bash scripts/setup.sh