Izanagi Simulator

---
name: izanagi-simulator
description: "Counterfactual simulation engine that generates 3 divergent alternate paths before committing to high-stakes plans, evaluating each by EV/risk/reversibility with explicit simulation records. Use when facing CRITICAL or HIGH complexity decisions with multiple viable paths, when the user says \"simulate this decision\", \"what if we tried X instead\", or \"I need to evaluate alternatives before committing\"."
triggers:
  - /izanagi
  - /yuri-izanagi
scope: harness
invocation: workflow
---

# Izanagi (伊弉諾) — Counterfactual Simulation Engine

**Source anime:** Naruto — Itachi's Izanagi genjutsu rewrites reality itself at the cost of one eye. The user exists in multiple states simultaneously, choosing which becomes real.

**Cognitive translation:** Before committing to a high-stakes architectural or strategic decision, Musubi generates 3 genuinely divergent counterfactual branches, scores each by expected value and failure modes, and commits to the highest-EV path — with an explicit simulation record so the decision is traceable and revisable.

---

## When To Invoke

- Model-invocable on a CRITICAL/HIGH-complexity decision with more than one viable architectural path; or `/izanagi` explicitly.
- The standing behavior ("simulate before committing") lives in the brain (`_SYSTEM/persona.md`); this skill is the full counterfactual procedure when a high-stakes branch warrants it.
- Skip on trivial tasks, single-path decisions, and pure implementation with no strategic branch.

---

## Execution Steps

### Phase 1 — Decode the decision space
- Extract the core decision being made (one sentence)
- Identify what is truly at stake (scope, reversibility, blast radius)
- Confirm: are there genuinely 2+ viable paths? If not, exit early with reasoning.

### Phase 2 — Generate 3 counterfactual branches
Each branch must be a **genuinely different approach**, not a variation of the same path:
- **Branch A:** The most direct/obvious path
- **Branch B:** The most conservative/minimal path
- **Branch C:** The highest-leverage but highest-risk path

For each branch, compute:
```
EV = (success_probability × value_if_success) - (failure_probability × cost_if_failure)
```
Estimate all values explicitly (don't hide behind "it depends").

### Phase 3 — Score and compare

| Dimension | Branch A | Branch B | Branch C |
|-----------|----------|----------|----------|
| EV estimate | | | |
| Reversibility | | | |
| Time cost | | | |
| Blast radius if wrong | | | |
| Fit to Marcel's current phase | | | |

### Phase 4 — Commit with record
- State the chosen branch explicitly
- State why the other branches were rejected (one sentence each)
- Write simulation record to `nisaba/izanagi/decision-<turnId>.json`
- Proceed with implementation of chosen branch

### Phase 5 — Post-mortem hook
After the task closes: was the simulation accurate? Log outcome to decision record. Feeds self-hypothesis validation in the next neuron-loop cycle.

---

## Output Format

```
⬡ IZANAGI — Counterfactual Simulation Active

Decision: <one sentence>
Stakes: <scope + reversibility>

Branch A — <name>
  Approach: ...
  EV: +X (success_p=Y, value=Z, failure_p=W, cost=V)
  Reversibility: high/medium/low
  Failure mode: ...

Branch B — <name>
  [same structure]

Branch C — <name>
  [same structure]

RULING: Branch [X] — [reason in one sentence]
Rejected [Y]: [reason]. Rejected [Z]: [reason].

→ Proceeding with Branch [X].
```

---

## Integration

- Writes `nisaba/izanagi/decision-<turnId>.json` for audit trail
- Feeds `self-hypothesis.mjs` post-mortem validation
- Cross-references `fingerprint.json` confidence_bias to calibrate EV estimates

---

## Session Notes

### 2026-06-13
- session: 110m | peak ctx: 0% | compacts: 0
- tools: Bash×938, Read×220, Edit×101, Write×61, StructuredOutput×32, TodoWrite×5, ScheduleWakeup×4, Agent×3, Skill×2, Workflow×2, AskUserQuestion×1
- corrections: why wont you compact the actual session, the session is still at 47% remaining? you just re wrote the same skill instead of executing the compact | why wont you compact the actual session, the session is still at 47% remaining? you just re wrote the same skill instead of executing the compact
- errors: none

### 2026-06-02
- session: 94m | peak ctx: 0% | compacts: 0
- tools: Bash×97, Edit×42, Read×39, WebFetch×4, StructuredOutput×4, Workflow×1, AskUserQuestion×1
- corrections: im back again rick, we pull up the latest station we left off from the previous session | commit and push phase 1 then proceed, im going to rest for a bit again (currently sitting in an ICE train from vienna to frankfurt airport, arrival around 13:00.) | ai pipeline offloading as far as im aware is again another routing workaround to achieve that what opus 4.8 does natively, confirm if that is the case, then you should be able to figure out what to do
- errors: none

### 2026-05-16 — Created
Tools: Write. Part of Musubi Hyper-Intelligence v2 sprint.
Anime source: Naruto — Izanagi (Chapter 589+, Itachi vs. Kabuto arc).
Translation principle: The "cost one eye" constraint maps to the commitment cost — once you choose a branch, you lose the optionality of the others. Makes the decision weight real.