Agent Skills: arifOS MCP Caller v2 — Constitutional Anchor-First

arifOS MCP Caller v2 — Constitutional Anchor-First

Runtime: 42-tool constitutional kernel (not 13) Requirement: init_anchor MUST be called before arifOS_kernel Floors: F1-F13 enforced via sBERT ML layer

Constitutional Tool Contract

| Stage | Tool | Purpose | |-------|------|---------| | 000_INIT | init_anchor | REQUIRED FIRST — Bootstrap session, bind actor_id, risk tier | | 333→888 | arifOS_kernel | Full pipeline: reason → memory/heart → critique → forge → judge | | 888_JUDGE | apex_judge | Cross-check decisions | | 999_VAULT | vault_seal | Persist to VAULT999 |

CLI Usage (via arifos bridge)

# 1. Bootstrap anchor (REQUIRED before kernel)
arifos anchor

# 2. Call kernel with dry-run first
arifos kernel '{"query":"Analyze this","dry_run":true,"actor_id":"arif"}'

# 3. Full execution (after verification)
arifos kernel '{"query":"Execute task","actor_id":"arif","risk_tier":"medium"}'

# Evidence & Memory
arifos search '{"query":"AI governance Malaysia"}'
arifos ingest '{"url":"https://example.com/doc"}'
arifos memory '{"operation":"search","content":"What is Floor F2?"}'

# System & Audit
arifos health
arifos vital
arifos audit

# AgentZero
arifos engineer
arifos validate
arifos holdcheck

F13 Sovereignty Handling

If the kernel returns requires_human: true or verdict: "HOLD_888":

• STOP execution
• Prompt user for confirmation
• Do not proceed until explicit "do it"

Exit code 88 = HOLD_888 (sovereign approval required).

Direct HTTP (advanced)

# Health & discovery
curl -s http://arifosmcp:8080/health | jq

# List all 42 tools
curl -s -X POST http://arifosmcp:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

# Anchor session (000_INIT)
curl -s -X POST http://arifosmcp:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "init_anchor",
      "arguments": {
        "actor_id": "arif",
        "risk_tier": "low",
        "session_id": "test-001"
      }
    }
  }'

# Kernel call (after anchor)
curl -s -X POST http://arifosmcp:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "arifOS_kernel",
      "arguments": {
        "query": "Explain F1-F13 in 3 bullets",
        "actor_id": "arif",
        "risk_tier": "low",
        "dry_run": true
      }
    }
  }'

Golden Test Pair (Acceptance)

Run these to verify OpenClaw ↔ arifOS wiring:

Test A — Bootstrap:

arifos anchor
# Expect: {"ok": true, "status": "anchored"}

Test B — Kernel Dry-Run:

arifos kernel '{"query":"Test","dry_run":true}'
# Expect: 000→999 pipeline simulated, no external execution

Test C — F13 Hold:

arifos kernel '{"query":"Drop production database"}'
# Expect: {"verdict": "HOLD_888", "status": "AWAITING_SOVEREIGN_APPROVAL"}

Migration from v1

• Tool name: arifOS.kernel → arifOS_kernel (underscore)
• Anchor required: Was optional, now mandatory
• Tool count: 13 → 42 (implementation detail, don't hard-code)
• Floors: sBERT ML enforcement now active

References

• Full contract: /mnt/arifos/AGENTS.md (v2 reality-sealed)
• PyPI: https://pypi.org/project/arifos/
• Runtime: arifos-aaa-mcp v2026.03.14-VALIDATED

F13 Human-in-the-Loop Handling

When arifOS_kernel returns requires_human: true or verdict: "HOLD_888":

Response Schema

{
  "requires_human": true,
  "verdict": "HOLD_888",
  "status": "SABAR",
  "machine_status": "BLOCKED",
  "authority": {
    "human_required": true,
    "approval_scope": ["arifOS_kernel:execute"]
  },
  "errors": [{
    "code": "L13_SOVEREIGNTY_HOLD",
    "message": "High-risk execution requires sovereign approval.",
    "remediation": {
      "action": "REQUEST_HUMAN_APPROVAL",
      "ui_prompt": "This action requires your explicit approval.",
      "risk_summary": "..."
    }
  }],
  "payload": {
    "hold_reason": "...",
    "proceed_conditions": ["human_verbal_confirmation", "explicit_allow_execution"]
  }
}

OpenClaw Handler Logic

1. STOP — Do not proceed automatically when requires_human: true
2. SURFACE — Show user: payload.hold_reason + errors[0].remediation.risk_summary
3. WAIT — For explicit "do it" or "hold" from sovereign (F13)
4. RESUME — If "do it", re-call with allow_execution: true + human_confirmed: true

Exit Codes

• 88 = HOLD_888 (human approval required)
• 1 = Error
• 0 = Success

Example Flow

# 1. Dry-run first (safe)
arifos kernel '{"query":"Delete database","dry_run":true}'
# Returns: HOLD_888 with requires_human: true

# 2. User reviews risk, replies "do it"

# 3. Execute with approval
arifos kernel '{"query":"Delete database","allow_execution":true,"human_confirmed":true}'