maxim

Maxim — the behavioral intelligence layer for Claude

The structural moat behind every Claude output. 91 specialist agents · 74 peer-reviewed behavioral frameworks · 14 compliance frameworks · 9 MCPs · 95 tools. Framework citation enforced on every output. Drift detected before you ship. Voice locked across every surface.

Install · then see the moat work in 30 seconds

/plugin marketplace add DrNabeelKhan/maxim
/plugin install maxim@maxim-packs

Then activate your tier (default: 90-day Trial of all 14 packs · no card · cancel anytime):

bash bootstrap/install-tier-packs.sh        # Mac · Linux · WSL · Git Bash
pwsh -File bootstrap/install-tier-packs.ps1 # Windows · or PS7 cross-platform

The wizard pre-selects Trial because a moat is hard to evaluate when you can't see it. Run your real work through Maxim for three months. Then decide what's worth keeping.

Per ADR-019 (Multi-Tenant Readiness): every operator starts with the same first-run experience. Pre-existing operators (like Maxim's maintainer running nk-writer) keep their advanced configurations untouched.

What Maxim actually IS

Most "AI plugins" give you templates. Some give you a chat wrapper. Maxim gives you a governed multi-agent operating system with structural enforcement of:

1. Framework citation on every output — every Maxim emission names the behavioral science framework that justifies it (Fogg · Cialdini · Prospect Theory · OWASP · NIST · 70 more). Outputs without citation get rejected pre-emit by behavioral-overlay-orchestrator (ADR-007). The "looks good" generic LLM output is structurally impossible to ship through Maxim.
2. Confidence tag on every output — every emission carries 🟢 HIGH / 🟡 MEDIUM / 🔴 LOW per ADR-010's Technical Educator rubric. You see grounding depth, not just an answer.
3. Compliance overlay that cannot be bypassed silently — 14 jurisdictional frameworks (GDPR · PIPEDA · UAE-PDPL · HIPAA · PCI-DSS · SOC2 · ISO 27001/13485/14971 · NIST CSF · EU AI Act · CASL · FINTRAC · WCAG 2.1) fire automatically on every regulated-data signal. CSO auto-loop is structural — even super-user mode doesn't disable it for compliance work.
4. Voice routing as a property of agent invocation — your voice loads when a writing agent dispatches, not when you remember to invoke /mxm-brand-voice first. Per ADR-016. Per operator via the v1.3.0 template pattern (ADR-019).
5. Proactive Watch drift detection — 13 universal drift classes scan every session start. Stale docs · broken refs · secret leaks · surface-claims drift · behavioral-moat drift. AI-coded projects rot fast; Watch is the rot detector.
6. Documents as Executable Contracts (ADR-002) — CHANGELOG · MOAT_TRACKER · BUG_TRACKER · AGENT_SKILL_INVENTORY are read by the pre-commit hook as live state. Drift between claim and reality blocks the commit.

None of the external sources (alirezarezvani · ui-ux-pro-max · superpowers · VoltAgent · planning-with-files) do any of this structurally. Maxim is the only layer that does. The community-pack system (ADR-008) composes Maxim's behavioral layer on top of their craft.

See the moat work · 6 concrete use cases

Use Case 1 · Founder evaluating regulatory exposure

You: "We're adding stablecoin payments to our pricing flow. What jurisdictions
      need DPAs and what's the controller/processor split?"

What fires: executive-router → cso-office → compliance-orchestrator + gdpr-counsel + hipaa-counsel + dpia-specialist. MCPs invoked: mxm-compliance.check_compliance scans 14 frameworks · mxm-compliance.generate_ropa_entry stubs the ROPA · mxm-compliance.get_jurisdiction_requirements per region.

What you get: jurisdictional map, cited per region, with DPA requirements, controller/processor split, and a flagged ROPA entry. 🟢 HIGH tag attached only if every jurisdiction in your compliance.frameworks returned PASS.

Without Maxim: 2-3 days research, $5K+ in consultant fees, no audit trail.

Use Case 2 · Engineer building a RAG pipeline with TDD discipline

You: "/mxm-build a RAG pipeline for our customer support docs"