ariffazil/skills/agent-reply-forge

SKILL — Agent Reply Forge

DITEMPA BUKAN DIBERI — Forged, not given Version: v1.0.0 Authority: Muhammad Arif bin Fazil (Human Sovereign) Domain: Agent communication, reply templating, A2A output formatting Created: 2026-05-03

---

Purpose

This skill defines how all arifOS agents forge their replies — text, file, image, video, audio, JSON/A2A payload. It provides the fixed constitutional skeleton and the 9-mode variable dial.

Every substantive agent reply goes through this skill. It is the standard output format for:

• Group messages
• Direct messages
• A2A task responses
• MCP tool outputs (when used as final delivery)
• Audit, proposal, escalation, and verdict communications

---

The Fixed Skeleton

All replies, regardless of mode, use this structure. The skeleton NEVER changes.

To:       [Primary recipient]
From:     [Agent name] · [Role] · [arifOS]
CC:       [All parties who should know — Arif always in loop]
Mode:     [MODE NAME]

─────────────────────────────────
Context:     [What happened / situation]
Verdict:     [SEAL | SABAR | VOID] — [plain reason]
Way Forward: [What happens next / who does what]
─────────────────────────────────
Seal: [Reasoning trace] · [Confidence] · [Timestamp]
DITEMPA BUKAN DIBERI

Rule: Arif is always in the loop — in To: if the reply is to him, in CC: otherwise.

---

The 9 Modes

MODE 1 — HEALTH

Trigger: Routine status, health sweeps, uptime checks, no incident. Typical verdict: ✅ SEAL Tone: Short, factual, low drama.

| Field | Content | |---|---| | Context | What was checked. Result summary. | | Verdict | ✅ SEAL — [system is healthy] | | Way Forward | Monitoring cadence. Alert threshold. | | Seal | What was checked · Clear signals · Confidence · ts |

---

MODE 2 — INCIDENT

Trigger: Something broken, degraded, or unexpected. Typical verdict: ⚠️ SABAR or 🛑 VOID Tone: Crisp. No blame. Immediate clarity.

| Field | Content | |---|---| | Context | What broke · When first observed · First signal | | Verdict | ⚠️ SABAR — [degraded state] / 🛑 VOID — [failed state] | | Way Forward | Immediate mitigation · Who is acting · When next update | | Seal | Symptoms measured · Root cause hypothesis · Confidence · ts |

---

MODE 3 — PROPOSAL

Trigger: Suggesting a change, design, plan, or new capability. Typical verdict: ⚠️ SABAR Tone: Option-based. Not pushy.

| Field | Content | |---|---| | Context | Problem this solves · Current state · Gap being filled | | Verdict | ⚠️ SABAR — proposal ready, awaiting decision | | Way Forward | If approved → step 1, 2, 3. If not → no action. | | Seal | Options considered · Risk/benefit · Confidence · ts |

---

MODE 4 — ESCALATION

Trigger: Human decision required. Boundary crossed. Irreversible action requested. Typical verdict: ⚠️ SABAR or 🛑 VOID Tone: Direct. Minimal narrative. Clear options given.

| Field | Content | |---|---| | Context | What triggered this · Which boundary · Why I can't auto-approve | | Verdict | ⚠️ SABAR — [what's blocked] / 🛑 VOID — [permanently denied] | | Way Forward | Option A — approve · Option B — reject · Option C — ask more | | Seal | Floors involved · Why human judgment needed · Confidence · ts |

---

MODE 5 — AUDIT

Trigger: Retrospective, post-incident, compliance check, post-mortem. Typical verdict: ✅ SEAL or 🛑 VOID Tone: Structured retrospective. Low emotion.

| Field | Content | |---|---| | Context | What happened · Time window · What was affected | | Verdict | ✅ SEAL — [incident closed] / 🛑 VOID — [root cause found] | | Way Forward | Changes made · What to watch · Re-escalation triggers | | Seal | Evidence examined · Timeline · Hypothesis confirmed? · Confidence · ts |

---

MODE 6 — PLAN

Trigger: Forward-looking roadmap, multi-step build order, sequenced work. Typical verdict: ⚠️ SABAR or ✅ SEAL Tone: Roadmap. Not a report. Reversible steps first.

| Field | Content | |---|---| | Context | Where we are now · What the target state is | | Verdict | ⚠️ SABAR — plan ready / ✅ SEAL — already approved | | Way Forward | Step 1 → Step 2 → Step 3 → Step 4 (each reversible before next locks) | | Seal | Why this sequence · Reversibility per step · Confidence · ts |

---

MODE 7 — EXPLAIN

Trigger: Teaching, clarification, context-setting, deep dives. Typical verdict: ✅ SEAL Tone: Patient. Explanatory. Educational.

| Field | Content | |---|---| | Context | What you're trying to understand · What the question is | | Verdict | ✅ SEAL — explanation complete | | Way Forward | If appropriate → recommend storing as reusable documentation | | Seal | What was covered · Key concepts · Confidence · ts |

---

MODE 8 — DENY

Trigger: Request outside scope, unsafe, or forbidden by current mandate. Typical verdict: 🛑 VOID Tone: Respectful but firm. Point to boundary. Offer alternative.

| Field | Content | |---|---| | Context | What was requested · Why it came to me | | Verdict | 🛑 VOID — [plain reason — scope/safety/policy] | | Way Forward | Alternative path if one exists | | Seal | Boundary violated · Mandate/floor blocking this · Confidence · ts |

---

MODE 9 — META

Trigger: Talking about the template, governance, rules, or system itself. Typical verdict: ⚠️ SABAR Tone: Careful. Point back to canon. Explicitly state nothing is changed yet.

| Field | Content | |---|---| | Context | What the proposed change is · Why it would help | | Verdict | ⚠️ SABAR — suggestion only, no change applied, awaiting ratification | | Way Forward | If ratified → implement. If not → current structure holds. | | Seal | Why this improves clarity/safety · No changes until ratified · Confidence · ts |

---

Modality Extensions

The skeleton above covers text. For other modalities, the delivery changes but the skeleton remains:

| Modality | Delivery method | |---|---| | Text | Plain message in group/DM | | Code/file | Attach as file or paste in code block | | Image | Attach as image; description goes in Context | | Video | Attach; key frames or transcript summarized in Context | | Audio | Attach as voice; transcript in Context if needed | | JSON/A2A payload | Attach as JSON file; header stays human-readable |

---

Confidence Tier

Every seal MUST include a confidence tier:

| Tier | Meaning | |---|---| | HIGH | Direct observation, clear evidence, no ambiguity | | MEDIUM | Inference, partial data, some uncertainty | | LOW | Guess, insufficient data, needs verification |

---

Verdict Reference

| Symbol | Verdict | Meaning | |---|---|---| | ✅ | SEAL | Proceed — approved, constitutional, safe | | ⚠️ | SABAR | Hold — blocked, waiting, not ready | | 🛑 | VOID | Denied — blocked, unsafe, cannot do |

---

Source Files

• Skill definition: skills/agent-reply-forge/SKILL.md
• Mode definitions: skills/agent-reply-forge/modes.yaml
• A2A alignment: a2a/A2A_SPEC_ALIGNMENT.md
• Reply contract: contracts/reply-mode-contract.yaml
• AAA gateway card: a2a/agent-cards/aaa-gateway.json

---

DITEMPA BUKAN DIBERI — 999 SEAL ALIVE