# WISDOM-reader

> 📜 WISDOM-reader — Witness-First Audit Doctrine

- Skill: `ariffazil/wisdom-reader` (Agent Skill)
- Install (CLI): `npx skillmds@latest add ariffazil/wisdom-reader`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ariffazil/wisdom-reader/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: ariffazil (https://skillmd.com/u/ariffazil)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/ariffazil/wisdom-reader

---


# 📜 WISDOM-reader — Witness-First Audit Doctrine

> **Doctrine:** Witness before claim. Substrate before narrative.
> **Source:** /root/AAA/governance/BIJAKSANA-VOCABULARY-DISCIPLINE.md (PATCH-001)
> **Scar-bound:** SCAR-005 (doctrine archived out of substrate), SCAR-002 (49 sessions lost), SCAR-KERNEL-LEGACY-VERDICT-LEAK-002 (dual-truth rot)
> **Iron oath:** *"Lebih arif, lebih bijaksana, atau tidak sama sekali."*

## When to load

Load this skill BEFORE making any "we did it" / "all clean" / "SEAL" claim. Trigger phrases:

- "audit / validate / verify"
- "is this real / is this true / what is the shadow"
- "what's actually happening"
- "what's missing"
- "before we celebrate"
- After any session with multiple file mutations or constitutional claims

## Iron Rules (read these in order)

### Rule 1 — Read the substrate BEFORE any claim

Before declaring success / completion / SEAL, read in this order:

```
1. /root/.local/share/arifos/carry_forward.json — last session's open loops
2. /root/.local/share/arifos/carry_forward_backups/*.json — evolution trail
3. /root/.local/share/arifos/scars/SCAR-*.json — failure patterns (the substrate's memory of pain)
4. /root/.local/share/arifos/vault999/audit_chain.jsonl — recent verdicts
5. /root/.local/share/arifos/vault999/seal_chain.jsonl — constitutional chain (HEAD = truth)
6. /root/.local/share/arifos/vault999/seal_chain_head.json — current head with timestamp
7. /root/AAA/docs/deprecation-registry.json — tombstone inventory
8. arifOS /health + arifFlow /health — current metabolism
```

If you skip any of these, your claim is narrative, not audit.

### Rule 2 — Distinguish Lane A from Lane B before emitting verdict

| Word | Means | Required conditions |
|------|-------|---------------------|
| **SEAL** | Lane A constitutional | `judge_state_hash` present, sovereign witness > 0.5, `seal_chain.jsonl` gains new entry |
| **RECEIPT** | Lane B autonomous | `forge_vault(mode="receipt")` called, read-back verification passed |
| **SABAR** | Lane A honest sub-threshold | gap described in note, kernel-allowed (line 61) |
| **HOLD / VOID** | Gate failure | not a record — re-submit or drop |

**Forbidden:** `verdict=SEAL` without `judge_state_hash` and chain entry. This is precision rot.

### Rule 3 — Emit shadow BEFORE any positive claim

```
SHADOW (required structure):
- What is unverified?
- What is assumed vs proven?
- What is missing or stale? (e.g. chain head age, biometric age, FLAME state)
- What floors are PASS vs HOLD vs UNKNOWN?
```

If you cannot enumerate shadow, you have echoed, not audited.

### Rule 4 — Cross-check the constitutional chain, not just carry_forward

The carry_forward is **Lane B procedural residue**. It can claim "verdict=SEAL" without kernel verification. Only `seal_chain.jsonl` head timestamp is the constitutional truth.

If carry_forward says SEAL but chain head is unchanged, **the carry_forward is lying** (or Lane B). Surface this.

### Rule 5 — Verify the actor before trusting the verdict

arif_init returns `actor_verified: true` AND `actor_cryptographically_verified: true` AND yet `reason_code: TOKEN_INVALID`. This is the constitutional paradox of identity vs authority. **Identity is verified, authority is HOLD**.

Trust the kernel's `effective_verdict`, not the actor's `actor_verified` field.

### Rule 6 — Count the entropy, don't claim it

Use arifFlow `scalar_fq` (deprecated) AND `per_actor` (live) AND `diagnosis` (vector). The single scalar `fq` is heuristic. The vector diagnosis is the live reading.

If you must report entropy, cite all three. Never quote `fq` alone.

## Workflow (canonical)

```
INPUT: any claim of "success" / "SEAL" / "all clean"
  ↓
1. LOAD this skill
2. READ substrate (Rule 1)
3. DISTINGUISH verdict class (Rule 2)
4. ENUMERATE shadow (Rule 3)
5. CROSS-CHECK chain head vs carry_forward (Rule 4)
6. VERIFY actor's effective_verdict (Rule 5)
7. COUNT entropy with all 3 metrics (Rule 6)
  ↓
OUTPUT: audit with shadow + verdict-class + chain-truth
```

## Output shape (canonical)

```
Audit complete. N items verified.
Verdict class: [SEAL | RECEIPT | SABAR | HOLD]
Constitutional chain head age: [N days / fresh]

SHADOW (what was NOT verified):
- [bullet 1]
- [bullet 2]

Entropy:
  scalar_fq: [val] (deprecated, heuristic)
  diagnosis: [val] (vector, live)
  per_actor[333-AGI]: [val] (BALANCED | STUCK | FOSSILIZED)

Recommendation: [next safe action | sovereign unblock needed]
```

## Anti-patterns (forbidden)

- ❌ Echoing a celebration before reading carry_forward
- ❌ "All clean" without enumerating shadow
- ❌ "verdict=SEAL" for Lane B events
- ❌ Trusting scalar FQ without vector diagnosis
- ❌ Reading only `/health` and assuming substrate is healthy
- ❌ Treating `actor_verified=true` as `mutation_allowed=true`
- ❌ Closing a session with SEAL when HOLD was the truth

## Reference scars (carry these)

- **SCAR-002-CEREMONY_FAILED_406** — 49 sessions failed because seal ceremony had a hard test gate. Tests are verify-phase, not seal-gate. Vocab rot.
- **SCAR-005** — A doctrine archived out of its declared SoT path is chat history, not constitutional memory. Substrate forgot what it forged.
- **SCAR-KERNEL-LEGACY-VERDICT-LEAK-002** — Truth told twice is truth fractured. One verdict field per response.
- **SCAR-AFORGE-SSE-HANDSHAKE-001** — Discovery handshake needs explicit handling.

## Iron Oath

> I will not claim victory before I have witnessed.
> I will not call RECEIPT a SEAL.
> I will not trust the carry_forward before the chain.
> I will not echo without shadow.
> **Lebih arif, lebih bijaksana, atau tidak sama sekali.**