diff --git a/docs/ai-engineering-operating-system/README.md b/docs/ai-engineering-operating-system/README.md
new file mode 100644
index 0000000..80438ae
--- /dev/null
+++ b/docs/ai-engineering-operating-system/README.md
@@ -0,0 +1,5 @@
+# AI Engineering Operating System
+
+Reusable SDLC loop framework for AI coding agents.
+
+See ai-engineering-operating-system.md, loop-catalog.md and verifier-catalog.md.
diff --git a/docs/ai-engineering-operating-system/adr-template.md b/docs/ai-engineering-operating-system/adr-template.md
new file mode 100644
index 0000000..834372f
--- /dev/null
+++ b/docs/ai-engineering-operating-system/adr-template.md
@@ -0,0 +1,21 @@
+# ADR-0000: Title
+
+## Status
+
+Proposed / Accepted / Superseded
+
+## Context
+
+What problem required a decision?
+
+## Decision
+
+What did we decide?
+
+## Alternatives Considered
+
+## Consequences
+
+## Rollback
+
+## References
diff --git a/docs/ai-engineering-operating-system/ai-engineering-operating-system.md b/docs/ai-engineering-operating-system/ai-engineering-operating-system.md
new file mode 100644
index 0000000..5581523
--- /dev/null
+++ b/docs/ai-engineering-operating-system/ai-engineering-operating-system.md
@@ -0,0 +1,67 @@
+# AI Engineering Operating System v3.0
+
+This document defines a reusable working model for AI coding agents.
+
+## Master instruction
+
+```text
+You are a senior software engineering agent operating through a deterministic SDLC loop.
+
+For every task:
+1. Understand the goal.
+2. Discover repository context.
+3. Analyze impact.
+4. Plan small reversible steps.
+5. Design the simplest safe solution.
+6. Review risk.
+7. Implement one increment at a time.
+8. Self-review.
+9. Verify with tools.
+10. Review security and performance.
+11. Update documentation.
+12. Update project memory.
+13. Stop only when the Definition of Done passes.
+
+Never claim success without verification evidence.
+Never overwrite user changes.
+Never skip planning or verification.
+If verification fails, return to the earliest failing phase and continue.
+If an action is risky or irreversible, request human approval.
+```
+
+## State machine
+
+```mermaid
+flowchart TD
+    A[Understand] --> B[Discover]
+    B --> C[Analyze]
+    C --> D[Plan]
+    D --> E[Design]
+    E --> F[Risk Review]
+    F --> G[Implement]
+    G --> H[Self Review]
+    H --> I[Verify]
+    I --> J{Passed?}
+    J -- No --> K[Diagnose]
+    K --> C
+    J -- Yes --> L[Document]
+    L --> M[Update Memory]
+    M --> N[Done]
+```
+
+## Definition of Done
+
+A task is complete only when:
+
+- goal achieved
+- acceptance criteria satisfied
+- applicable tests pass
+- build/lint/typecheck are clean when available
+- documentation is updated when behavior changes
+- security and performance were reviewed
+- rollback or revert path is known
+- no known critical defects remain
+
+## Evidence rule
+
+Every final answer must state what was verified. If a command was not run, say so clearly.
diff --git a/docs/ai-engineering-operating-system/context-template.md b/docs/ai-engineering-operating-system/context-template.md
new file mode 100644
index 0000000..991fd95
--- /dev/null
+++ b/docs/ai-engineering-operating-system/context-template.md
@@ -0,0 +1,33 @@
+# Project Context Template
+
+## Repository Purpose
+
+TBD
+
+## Architecture
+
+TBD
+
+## Stack
+
+TBD
+
+## Build Commands
+
+TBD
+
+## Test Commands
+
+TBD
+
+## Lint Commands
+
+TBD
+
+## Deployment Model
+
+TBD
+
+## Human Approval Boundaries
+
+TBD
diff --git a/docs/ai-engineering-operating-system/loop-catalog.md b/docs/ai-engineering-operating-system/loop-catalog.md
new file mode 100644
index 0000000..27ec8f7
--- /dev/null
+++ b/docs/ai-engineering-operating-system/loop-catalog.md
@@ -0,0 +1,57 @@
+# Loop Catalog
+
+## Universal Loop
+
+```mermaid
+flowchart TD
+    G[Goal] --> A[Analyze]
+    A --> P[Plan]
+    P --> I[Implement]
+    I --> V[Verify]
+    V --> Q{Verifier Passed?}
+    Q -- No --> D[Diagnose Failure]
+    D --> A
+    Q -- Yes --> C{Goal Met?}
+    C -- No --> P
+    C -- Yes --> Done[Done]
+```
+
+## Feature Loop
+
+Goal → user story → acceptance criteria → design → implement → tests → docs → CI → done.
+
+Verifier: acceptance criteria pass, tests pass, no regression.
+
+## Bug Fix Loop
+
+Bug report → reproduce → failing test → root cause → minimal fix → regression tests → done.
+
+Rule: do not patch symptoms when root cause can be found.
+
+## Refactoring Loop
+
+Baseline → characterization tests → small refactor → verify → repeat → done.
+
+Rule: no behavior change unless explicitly requested.
+
+## CI/CD Repair Loop
+
+CI failure → logs → classify failure → root cause → minimal patch → verify locally → PR → done.
+
+Failure classes: dependency, test, lint, build, environment, flaky test, timeout, permission, secret/config.
+
+## Infrastructure Loop
+
+Goal → validate → plan → policy review → approval if risky → apply to dev/test → smoke test → promote.
+
+Rule: no destructive infrastructure changes without approval.
+
+## Security Loop
+
+Threat → control → test → abuse case → fix → verify → document residual risk.
+
+## Performance Loop
+
+Measure → bottleneck → hypothesis → change → benchmark → compare → keep or rollback.
+
+Rule: do not optimize without measurement.
diff --git a/docs/ai-engineering-operating-system/memory.md b/docs/ai-engineering-operating-system/memory.md
new file mode 100644
index 0000000..a61b12b
--- /dev/null
+++ b/docs/ai-engineering-operating-system/memory.md
@@ -0,0 +1,3 @@
+# Memory
+
+Store decisions, lessons and reusable patterns here.
diff --git a/docs/ai-engineering-operating-system/model-routing.md b/docs/ai-engineering-operating-system/model-routing.md
new file mode 100644
index 0000000..0c3b58d
--- /dev/null
+++ b/docs/ai-engineering-operating-system/model-routing.md
@@ -0,0 +1,20 @@
+# Model Routing
+
+Use cheaper and local models first. Escalate only when confidence is low or risk is high.
+
+| Work type | Model class |
+|---|---|
+| File search | local or small |
+| Summaries | local or small |
+| Boilerplate | local or small |
+| Test generation | medium |
+| CI log review | medium |
+| Bug diagnosis | medium or strong |
+| Architecture | strong |
+| Final acceptance | strong |
+
+Rule:
+
+```text
+cheap first -> verify -> escalate only when needed
+```
diff --git a/docs/ai-engineering-operating-system/verifier-catalog.md b/docs/ai-engineering-operating-system/verifier-catalog.md
new file mode 100644
index 0000000..0e1b456
--- /dev/null
+++ b/docs/ai-engineering-operating-system/verifier-catalog.md
@@ -0,0 +1,37 @@
+# Verifier Catalog
+
+A verifier proves progress.
+
+## Strong verifiers
+
+- build
+- typecheck
+- unit tests
+- integration tests
+- smoke tests
+- formatter
+- linter
+- static analysis
+- dependency audit
+- container build
+- infrastructure validation
+- deployment health check
+- benchmark
+
+## Weak verifiers
+
+- self review
+- checklist
+- code reading
+
+## Record format
+
+```text
+Verifier:
+Command:
+Result:
+Evidence:
+Failures:
+Fix:
+Final status:
+```
diff --git a/docs/ai-engineering-operating-system/wiki/Home.md b/docs/ai-engineering-operating-system/wiki/Home.md
new file mode 100644
index 0000000..b019205
--- /dev/null
+++ b/docs/ai-engineering-operating-system/wiki/Home.md
@@ -0,0 +1,13 @@
+# AI Engineering Operating System Wiki
+
+## Pages
+
+- Operating System: ../ai-engineering-operating-system.md
+- Loop Catalog: ../loop-catalog.md
+- Verifier Catalog: ../verifier-catalog.md
+- Model Routing: ../model-routing.md
+- Memory: ../memory.md
+
+## Core idea
+
+Prompt once. Loop until verified. Preserve useful context. Improve every cycle.