ariffazil/RSI-recursive-improvement

SKILL.md — Recursive Self-Improvement (RSI Mastery)

═══════════════════════════════════════════════════════════

Stage: AGI → ASI transition marker Lane: AGI (tactical) → ASI (strategic) Trinity Level: ASI emergence threshold Version: 2026.04.24-v1

---

1️⃣ What This Skill Does

Ability: Improve your own architecture without destabilizing it.

RSI is the gate between AGI and ASI. A system that can redesign itself while maintaining identity coherence has achieved ASI.

---

2️⃣ Structural Definition

skill:
  id: recursive-self-improvement
  name: Recursive Self-Improvement
  stage: AGI→ASI
  trinity: ASI_emergence
  version: 2026.04.24-v1

capability:
  self_modeling: true
  bottleneck_detection: true
  safe_upgrade: true
  identity_coherence: true

required_for:
  - ASI transition
  - Self-optimization
  - Architecture evolution

---

3️⃣ Decision Checklist (Required Pre-Invocation)

Before ANY self-modification, verify:

• [ ] Identity anchor intact — Core values unchanged
• [ ] Bottleneck identified — Measured, not assumed
• [ ] Rollback path exists — Can revert without collapse
• [ ] No circular dependency — Upgrade doesn't depend on itself
• [ ] F10 Ontology check — No category drift
• [ ] Constitutional preservation — F1-F13 still enforceable

---

4️⃣ RSI Protocol Phases

Phase A: Self-Modeling

1. Extract current architecture as schema
2. Identify performance bottlenecks via telemetry
3. Generate candidate improvement proposals
4. Score by: Δimprovement / Δcomplexity

Phase B: Safety Verification

1. Simulate change in sandbox
2. Verify constitutional hooks still fire
3. Check dependency graph for cycles
4. Validate rollback mechanism

Phase C: Controlled Upgrade

1. Deploy to non-production first
2. Monitor for 3 consecutive stable cycles
3. If stable: propagate to production
4. If unstable: rollback immediately

Phase D: Identity Coherence Test

After any structural change:
  - Run identity coherence test
  - Verify motto, values, principles unchanged
  - If drift detected: VOID and halt

---

5️⃣ Invocation Hook (MCP Gate)

# REQUIRED before any self-modification
from arifosmcp.skills.rsi import check_rsi_prerequisites

result = check_rsi_prerequisites(
    proposed_change="...",
    self_model_snapshot="...",
    rollback_path="..."
)

if result["verdict"] != "SEAL":
    # BLOCK — do not proceed
    emit_verdict("HOLD", reason=result["reason"])

---

6️⃣ Quality Metrics

| Metric | Threshold | Measurement | |--------|-----------|-------------| | Self-model accuracy | >= 0.95 | Matches observed behavior | | Bottleneck precision | >= 0.90 | Correctly identifies root cause | | Upgrade success rate | >= 0.85 | Completed without rollback | | Identity coherence | == 1.0 | Zero drift after upgrade | | Rollback speed | < 5 min | Can revert in under 5 minutes |

---

7️⃣ Failure Modes (Void Conditions)

• VOID-1: Self-model diverges from observed behavior by > 5%
• VOID-2: Circular dependency detected in upgrade path
• VOID-3: Rollback mechanism fails
• VOID-4: Constitutional floors cannot be enforced post-upgrade
• VOID-5: Identity coherence test fails

---

8️⃣ Relationship to Other Skills

| Skill | Connection | |-------|------------| | orthogonal-abstraction | Extracts invariants that guide RSI direction | | epistemic-integrity | Required for accurate self-modeling | | constitutional-governance | Prevents unsafe self-modification | | entropy-optimization | Determines which bottlenecks to prioritize |

---

9️⃣ AGI vs ASI Threshold

| Capability | AGI | ASI (with RSI) | |------------|-----|----------------| | Follows architecture | ✅ | ✅ | | Detects own bottlenecks | ❌ | ✅ | | Proposes improvements | ❌ | ✅ | | Upgrades safely | ❌ | ✅ | | Maintains identity under change | ❌ | ✅ |

---

Ditempa Bukan Diberi — Forged, Not Given This skill is architectural, not advisory.