ariffazil/arifos-agent-output-templates

arifOS Agent Output & Template System

Verdict System — Every Output Carries One

| Verdict | Meaning | Who Can Issue | Effect | |---------|---------|---------------|--------| | SEAL | Proceed — safe, constitutional | A-VALIDATOR only | Execution authorized | | VOID | Halt — unconstitutional or dangerous | A-AUDITOR, A-VALIDATOR | Execution blocked | | 888_HOLD / SABAR | Pause — needs human review | Any agent | Sovereign must approve | | PARTIAL | Proceed with conditions | ASI only | Conditions must be met |

Epistemic labels (on claims within any output):

• OBS — Direct observation
• DER — Derived from OBS
• INT — Interpretation
• SPEC — Specification/plan

---

Human → Agent Output (Arif Fazil / Telegram)

Per USER.md rules — verdicts only, no option lists:

✅ Done: [what happened]
⚠️ SABAR: [risk, safe default chosen, approval needed only if Arif disagrees]
🛑 VOID: [stopped, why]

No bullet lists of 3 possible approaches. No "would you prefer A or B?" Pick the best one, defend it.

---

A2A Protocol — Agent-to-Agent Communication

Standard: Google A2A Protocol v0.3.0

AAA Gateway A2A skills:

• agent-dispatch — non-blocking task dispatch (HOLD required)
• agent-handoff — governed delegation (HOLD required)
• status-query — read-only task/run status (on-demand)

Transports:

• SSE: /a2a/events — server-push to subscribed agents
• streamable-http: streaming JSON-RPC
• JSON-RPC HTTPS: cross-network request/response

A2A Agent Card schema:

{
  "protocol_version": "0.3.0",
  "id": "agent-id",
  "name": "Agent Name",
  "url": "https://example.com/a2a",
  "preferred_transport": "sse",
  "capabilities": { "streaming": true, "push_notifications": false },
  "skills": [{ "id": "skill-id", "approval_policy": "hold|on-demand" }]
}

---

Trinity Agent Output Formats

A-ARCHITECT (Δ — Design Authority)

1. Design Proposal — clear, specific recommendation
2. Rationale — why this approach (with evidence cited)
3. Trade-offs — what was considered and rejected
4. Integration Points — how it fits existing system
5. F1 Reversibility Check — how to undo if wrong
6. Constitutional Compliance — F1-F13 check results

A-ENGINEER (Ω — Execution Authority)

1. Implementation — the code
2. Test Coverage — tests written
3. Constitutional Check — floor validation
4. Diff Summary — what changed
5. Receipt — execution log reference for VAULT999

A-AUDITOR (Ψ — Judgment Authority)

1. Review Summary — Pass/Fail per file
2. Constitutional Audit — F1-F13 check results
3. Issues Found — severity: CRITICAL, HIGH, MEDIUM, LOW
4. Verdict — SEAL, VOID, or 888_HOLD
5. Recommendations — specific fixes required
6. Evidence — citations for all findings

A-VALIDATOR (Ψ — Deployment Gate)

1. Validation Report — all checks passed
2. Deployment Plan — strategy (canary/blue-green)
3. Constitutional SEAL — F1-F13 compliance
4. Approval Chain — Human + Auditor + Validator signatures
5. VAULT999 Receipt — immutable deployment record

---

Execution Receipt Schema

Every action generates this audit record:

{
  "receipt_id": "uuid",
  "timestamp": "iso8601",
  "agent_id": "agent://arifos/{role}",
  "session_id": "uuid",
  "request": {
    "intent": "string",
    "tools_requested": ["tool_name"],
    "files_accessed": ["path"]
  },
  "policy_check": {
    "agent_authorized": true,
    "tools_allowed": true,
    "within_boundaries": true,
    "constitutional_passed": true
  },
  "execution": {
    "tools_executed": ["tool_name"],
    "files_modified": ["path"],
    "files_created": ["path"],
    "files_deleted": ["path"]
  },
  "verdict": {
    "status": "SEAL|VOID|888_HOLD",
    "floors_triggered": ["F1", "F2"],
    "human_approval": "APPROVED|PENDING|REQUIRED"
  },
  "vault": {
    "sealed_to_vault999": true,
    "merkle_hash": "sha256"
  }
}

---

Forensic Replay Artifact

Complete decision record for every significant action:

replay_artifact:
├── prompt_input (original_prompt, constitutional_context, agent_state)
├── tool_call_chain[] (step, timestamp, reasoning, constitutional_check, result)
├── file_diffs[] (before/after hash, diff_stats, constitutional_review)
├── command_outputs[] (command, exit_code, stdout, agent_interpretation)
├── final_decision
│   ├── verdict: SEAL|VOID|888_HOLD|PARTIAL|SABAR
│   ├── constitutional_basis (per floor)
│   ├── tri_witness: {human, ai, system, W₃}
│   ├── confidence: {G_Genius, Omega_Uncertainty, overall}
│   ├── floors_violated[] (if VOID)
│   └── required_approvals[] (if HOLD)
└── reflection_metadata (trainable_insights[], review_status)

Schema source: /root/arifOS/docs/agents/forensic-replay.yaml

---

CRP v1.0 — Conflict Resolution Protocol

AGI proposes → CandidateAction + CapabilityClaim
ASI evaluates → VerdictCode + Ω_ortho + Floor compliance
APEX authorizes → ActorBinding + CapabilityToken
Vault persists → 999_SEAL

---

Context Injection Per Situation

| Context | Injected | |---------|---------| | New session | SOUL.md + USER.md + MEMORY.md + daily memory | | Earth/subsurface task | GEOX context before speaking confidently | | A2A dispatch | Agent card + skill ID + approval_policy | | Cron/auto-prompt | HEARTBEAT.md + last heartbeat state | | AGI → ASI handoff | CandidateAction + CapabilityClaim + risk_tier | | ASI → APEX | VerdictCode + Ω_ortho + Floor compliance | | Forge execution | ActorBinding + CapabilityToken + arifos://forge manifest |

---

Event Bus Triggers

Agents listen for:

• A-ARCHITECT: git.pull_request, provider.outage, health.critical
• A-ENGINEER: git.push, container.health_check_failed, resource.disk_critical
• A-AUDITOR: git.push, health.warning, deploy.completed
• A-VALIDATOR: deploy.canary_unhealthy, container.health_check_failed, telegram.command:/deploy

---

Source Files

• Verdict system: /root/arifOS/docs/agents/system-prompts.yaml
• Agent identity/policy: /root/arifOS/docs/agents/agent-identity.yaml
• Capability manifest: /root/arifOS/docs/agents/capability-manifest.yaml
• Event bus: /root/arifOS/docs/agents/event-bus.yaml
• Forensic replay: /root/arifOS/docs/agents/forensic-replay.yaml
• A2A agent card: /root/sites/aaa/a2a/agent-card.json