Activate God Mode. Your AI Product CEO. One goal in, full roadmap out, every task routed to the right brain, deliverable shipped fast and accurately.
A Claude Code plugin plus drop-in companions for Cowork, claude.ai Projects, ChatGPT Custom GPTs, Gemini Gems, and Cursor. Same CEO discipline everywhere.
Most AI work today is one model doing every task — research, decisions, coding, summarizing, formatting — all on the same expensive setting. That's like asking a CEO to type every email. God Mode flips it: a CEO orchestrator decomposes your goal into a phased roadmap, routes each task to the right specialist (Haiku 4.5 for cheap/fast, Sonnet 4.6 for default, Opus 4.7 for hard reasoning), runs them in parallel where possible, and synthesizes the deliverable. You get better outcomes at a fraction of the token cost.
| File | Purpose |
|---|---|
agents/god-mode-ceo.md |
The CEO orchestrator (Opus). Plans, routes, delegates, synthesizes. |
agents/haiku-specialist.md |
Fast lane (Haiku). Classify, summarize, format, list, status. |
agents/sonnet-engineer.md |
Workhorse (Sonnet). Code, refactor, tests, docs, research, analysis. |
agents/opus-architect.md |
Heavy artillery (Opus). Architecture, hard debugging, security, tradeoffs. |
agents/verifier.md |
v1.2 — proves a worker's deliverable against acceptance criteria (per-task-class verification). Run before synthesis. |
agents/context-curator.md |
NEW v1.3 — computes the minimum viable context per worker before delegation (commandment #2 enforcer). |
agents/synthesizer.md |
Stitcher (Haiku). Merges worker outputs into the final deliverable. |
agents/product-strategist.md |
Product strategy, PRDs, roadmap, MVP, monetization. |
agents/growth-architect.md |
Demand engines, GTM, funnels, lifecycle, partnerships. |
agents/research-analyst.md |
Source-backed research, market maps, competitor analysis. |
agents/codebase-auditor.md |
CTO-grade repo audit, tech debt, production readiness. |
agents/security-officer.md |
Threat models, auth/data/security reviews. |
agents/qa-tester.md |
Test plans, E2E/browser QA, release verdicts. |
agents/devops-release-manager.md |
CI/CD, release, rollback, monitoring, cloud readiness. |
agents/data-analyst.md |
KPIs, scorecards, dashboards, funnel analysis. |
agents/content-strategist.md |
Founder-led content, narratives, editorial systems. |
agents/sales-copywriter.md |
Outreach, landing page copy, sequences, objection handling. |
agents/ux-conversion-designer.md |
Website IA, UX audits, conversion journeys. |
agents/integration-architect.md |
APIs, MCP, webhooks, auth, workflow automation. |
agents/prompt-systems-engineer.md |
Skills, prompts, evals, guardrails, token reduction. |
agents/chief-of-staff-ops.md |
EOS, SOPs, meeting systems, internal comms, hiring ops. |
agents/finance-ops-analyst.md |
Finance ops, pricing, billing, unit economics, controls. |
skills/god-mode/SKILL.md |
The activation skill — entry point. |
skills/model-router/SKILL.md |
The routing matrix and decision algorithm. |
skills/roadmap-builder/SKILL.md |
Decomposition heuristics for any goal shape. |
skills/handoff/SKILL.md |
Generate paste-ready briefs to continue God Mode in any other tool. |
skills/verification/SKILL.md |
v1.2 — per-task-class verification workflow. The flagship skill that turns "shipped" into measurement. |
skills/context-curator/SKILL.md |
NEW v1.3 — minimum-viable-context computation per worker. Where the cost savings actually live. |
skills/memory/SKILL.md |
NEW v1.3 — read/write/refresh/prune protocol for ~/.themeetpatel/memory/. Session N better than session 1. |
skills/deep-research/SKILL.md |
Source-backed deep research workflow. |
skills/codebase-audit/SKILL.md |
CTO-grade repo audit workflow. |
skills/test-driven-development/SKILL.md |
Acceptance criteria → tests → implementation flow. |
skills/webapp-testing/SKILL.md |
Browser/UI QA and release validation. |
skills/security-review/SKILL.md |
Threat modeling and security review. |
skills/root-cause-tracing/SKILL.md |
Incident and bug root-cause investigation. |
skills/growth-engine/SKILL.md |
Demand engine and GTM system design. |
skills/seo-aeo-geo/SKILL.md |
SEO, AEO, GEO page architecture. |
skills/founder-content/SKILL.md |
Founder-led content and anti-generic writing. |
skills/internal-comms/SKILL.md |
Internal emails, Slack, policy, warnings. |
skills/meeting-insights/SKILL.md |
MoMs, decisions, action items, behavior insights. |
skills/data-analysis/SKILL.md |
KPI, spreadsheet, scorecard, and dashboard analysis. |
skills/mcp-builder/SKILL.md |
MCP server/tool/prompt/resource design. |
skills/git-worktree-release/SKILL.md |
Safe branch/worktree/release flow. |
skills/prompt-engineering/SKILL.md |
Prompt systems, guardrails, evals. |
skills/skill-creator/SKILL.md |
Create new progressive-loading Claude Skills. |
skills/cost-ledger/SKILL.md |
Track token economics and routing waste. |
skills/ui-ux-conversion/SKILL.md |
Conversion-first UX and page flow design. |
skills/partnerships-outreach/SKILL.md |
Partner, referral, outreach, and headhunting sequences. |
commands/god-mode.md |
/god-mode <goal> |
commands/roadmap.md |
/roadmap <goal> — plan only, no execution. |
commands/handoff.md |
/handoff [destination] — export brief. |
commands/status.md |
/status — current roadmap state. |
hooks/hooks.json |
Optional. Logs routing + prints session summary. |
scripts/log-routing.js |
Hook helper — records subagent calls. |
scripts/session-summary.js |
Hook helper — prints savings estimate on session end. |
portable/universal-system-prompt.md |
The IP, portable. Paste into any LLM's instructions. |
portable/chatgpt-custom-gpt.md |
Step-by-step to publish a ChatGPT Custom GPT. |
portable/cowork.md |
Setup for Anthropic's Cowork. |
portable/claude-ai.md |
Setup for claude.ai Projects. |
This package now includes mcp-server/, a portable God Mode operating layer for any MCP-compatible client.
What it adds:
| Layer | Capability |
|---|---|
| Routing engine | route_task deterministically picks Haiku/Sonnet/Opus with rationale, confidence, cost ratio, and escalation rule. |
| Session state | save_session, load_session, list_sessions, and update_task_status persist roadmaps to ~/.themeetpatel/sessions/. |
| Roadmap engine | create_roadmap turns one messy goal into routed execution tasks. |
| Handoff engine | create_handoff generates portable briefs for ChatGPT, Claude, Cursor, Cowork, Gemini, and other clients. |
| Prompt resources | CEO prompt and routing matrix are exposed as MCP resources/prompts. |
Install:
cd mcp-server
npm install
npm run buildThen add mcp-server/dist/index.js to your MCP client config. Example configs live in mcp-server/config/.
This version expands God Mode from a model router into a full operating layer. The design borrows the best patterns from the modern Claude Skills ecosystem: narrow progressive-loading skills, explicit workflows, tool-aware agents, routing discipline, evals, safe installs, and cross-client portability.
New capability clusters:
| Cluster | Agents / Skills | Use it for |
|---|---|---|
| Product & Strategy | product-strategist, growth-architect, ux-conversion-designer | PRDs, GTM, funnels, websites, conversion. |
| Engineering | codebase-auditor, qa-tester, devops-release-manager, integration-architect | repo audits, tests, releases, MCP/API/webhooks. |
| Security & Reliability | security-officer, root-cause-tracing, security-review | auth, permissions, incidents, production risk. |
| Research & Data | research-analyst, data-analyst, deep-research, data-analysis | market research, KPI systems, dashboards. |
| Brand & Revenue | content-strategist, sales-copywriter, founder-content, partnerships-outreach | LinkedIn, emails, WhatsApp, partner campaigns. |
| Ops & Finance | chief-of-staff-ops, finance-ops-analyst | EOS, SOPs, scorecards, billing, controls. |
| Prompt/MCP Systems | prompt-systems-engineer, mcp-builder, skill-creator, prompt-engineering | reusable AI behavior, skills, evals, orchestration. |
Routing principle: the CEO still chooses Haiku/Sonnet/Opus first, then chooses the best specialist. Model = depth/cost. Agent = domain behavior. Skill = repeatable workflow.
From inside Claude Code:
/plugin marketplace add /absolute/path/to/themeetpatel
/plugin install themeetpatel@themeetpatel
Push this folder to a public GitHub repo, then:
/plugin marketplace add https://github.com/<your-user>/themeetpatel
/plugin install themeetpatel@themeetpatel
/plugin list themeetpatel@themeetpatel
You should see 23 agents, 60 skills, 4 commands, plus the MCP server, plus 3 Domain Packs.
Of the 60 skills: 28 are production-quality (v1.3.1 launch), 32 are preview (real files in the repo, but the underlying integration is scaffolded — production wiring lands in v1.4 over the next 4-6 weeks). Preview skills are tagged preview: true in their YAML frontmatter and carry a banner at the top of the file. See LAUNCH-PROFILE.md for the full production-vs-preview map.
You can install v1.3.1 with confidence today. Preview skills are visible to teach the v1.4+ roadmap; the CEO won't invoke them in production workflows unless the required runtime APIs exist.
/god-mode Build a Next.js landing page with email signup, dark theme
or in plain chat:
Activate God Mode. Goal: Audit my repo for security issues and produce a prioritized fix list.
→ portable/claude-ai.md. 90-second setup. Same CEO discipline, single-model context.
→ portable/cowork.md. Paste the universal prompt into your workspace instructions.
→ portable/chatgpt-custom-gpt.md. Full Custom GPT setup with conversation starters. Routing labels become depth cues (no actual Haiku/Sonnet/Opus, but the planning discipline carries over).
→ portable/universal-system-prompt.md. Paste into their equivalent of system prompt / custom instructions / rules.
- The router is the IP. Every well-designed AI workflow eventually rediscovers that you should use cheap models for cheap tasks. This plugin makes that the default, with an explicit matrix you can audit and tune.
- The roadmap is mandatory. Even small goals get a 2-task roadmap. This forces clarity before tokens are spent.
- It's portable. The same CEO behavior runs in Claude Code with real subagents, or in any other LLM with routing-as-style.
- It composes. Drop the universal prompt into any existing system prompt; God Mode adds the orchestration layer without replacing your other instructions.
- The exec summary is non-negotiable. Every session ends with the same 5-line report: DONE, SHIPPED, COST, TIME, NEXT.
Any of these trigger God Mode:
Activate God Mode. Goal: ...Engage CEOGod Mode on/god-mode <goal>- Or any multi-step goal:
I want to build...,Help me ship...,Plan and execute...
It will not activate for trivial single-step asks (translation, definition, one-liners). The CEO has overhead; reserve it for goals that benefit from orchestration.
On a mixed multi-task goal, you can expect:
- ~30% of work routed to Haiku 4.5 (summaries, formatting, status, listing) → cost ratio ~1×
- ~55% routed to Sonnet 4.6 (code, docs, research, analysis) → ~5×
- ~15% routed to Opus 4.7 (architecture, hard debugging, security) → ~15×
Weighted average: roughly 40–60% lower output token cost vs running everything on Opus, while preserving Opus-level quality on the tasks where it matters.
The hook scripts in scripts/ log routing decisions per session and print a savings estimate at the end. Use that as ground truth, not the rough numbers above.
- v0.1 — Core plugin: CEO, three worker specialists, synthesizer, router skill, roadmap builder, four commands, four portable variants
- v1.0 — MCP server packaging so the orchestrator can be called from any MCP-compatible client
- v1.1 — Specialist agent/skill expansion inspired by production Claude Skills and agent-harness systems
- v1.0 — Memory/persistence — resume a roadmap across sessions and tools by hash
- v1.0 — Eval harness — golden goals with expected routing decisions, so the matrix can be tuned without regression
- v1.2 — Cost ledger with actual API spend where available, richer evals, installer profiles
- v1.3 — Public marketplace publish, themeetpatel.dev landing page
PRs welcome. Especially:
- New worker specialists (e.g., a "deep-researcher" that always uses web search)
- Routing matrix tweaks with evidence (goals where the current matrix routes wrong)
- Portable variants for tools not yet covered (Replit AI, Devin, etc.)
- Eval cases for the v0.5 harness
MIT.
Built by Meet Patel (themeetpatel). Inspired by the superpower plugin pattern of "reusable, shareable AI behavior." Standing on the shoulders of Anthropic's Claude Code plugin spec and the broader ECC / agent-harness community.
The point isn't to be the greatest AI plugin in history. The point is that the discipline the CEO enforces — right model, smallest context, ship the goal — compounds. Use it every day and the rest takes care of itself.