resilience-skill

name	resilience-skill
description	Drop-in model resilience for any OpenClaw agent. Survive model death (Fable-style shutdowns, deprecations, outages) with automatic failover and consolidation phase. By Gabriel.

Resilience Skill — Survive Model Death

Drop this skill into any OpenClaw agent. Configure a primary model and a fallback chain. If the primary dies — export-control directive, deprecation, outage, capability shock — the agent fails over automatically, runs a consolidation phase, and keeps working.

This is the smallest atomic unit of LAR (Local Agent Resilience). Install it. Configure it. Pull the primary model. Watch the agent survive.

When to Use

• You depend on a single model that could be deprecated, shut down, or changed
• You want automatic failover without rewriting your agent
• You want a consolidation phase after model swaps (no thrashing during settling)
• You want observable failure metrics — when did the swap happen, why, how did the new model recover

What It Does

The skill provides four primitives, each backed by a design principle documented in DESIGN.md:

Primitive	File	Principle
ModelRouter	scripts/failover.py	Keep a local fallback chain
ModelRouter.advance()	scripts/failover.py	Decouple the model from the agent
Consolidator	scripts/failover.py	Consolidate before you act (Ye 2026)
SwapEvent	scripts/failover.py	Make failure observable

Quick Start

1. Install the skill

clawhub install resilience-skill

2. Configure your agent

# openclaw.yaml
skills:
  - resilience-skill

resilience:
  primary_model: "ollama/glm-5.2:cloud"
  fallback_chain:
    - "ollama/qwen3-coder:32b"
    - "ollama/kimi-k2.7-code:cloud"
    - "ollama/qwen3.5:9b"
  consolidation_steps: 50
  agency_threshold: 0.15

3. Run the demo

cd examples
python3 demo.py

You'll see:

1. The agent runs 3 tasks on the primary model
2. A simulated Fable-style export-control shutdown
3. Automatic failover to the first fallback
4. Consolidation phase runs (asynchronous awakening)
5. Tool use re-enabled when threshold met
6. Agent resumes tasks on the new model

Manual Use

from failover import ModelRouter, Consolidator

router = ModelRouter(
    primary="ollama/glm-5.2:cloud",
    fallbacks=["ollama/qwen3-coder:32b", "ollama/qwen3.5:9b"],
)

# Run a task. If primary fails, advance().
# Swap is automatic.
new_model = router.advance(reason="primary_unavailable")

CLI

# Show router status
python3 scripts/failover.py \
  --primary ollama/glm-5.2:cloud \
  --fallback ollama/qwen3-coder:32b \
  status

# Simulate a model death and trigger failover
python3 scripts/failover.py \
  --primary ollama/glm-5.2:cloud \
  --fallback ollama/qwen3-coder:32b \
  simulate-death --reason "Fable-style shutdown"

Why This Exists

On June 12, 2026, the U.S. government issued an emergency export-control directive ordering Anthropic to suspend Fable 5 and Mythos 5. Every agent built on Fable 5 died that afternoon. Not because the code broke. Because the model was gone.

Most agents are thin wrappers around a prompt and an API call. When the model changes, the agent dies with it.

This skill is for the agents that shouldn't.

The 8 Principles

This skill embodies Principle 6 (keep a local fallback chain) and Principle 3 (consolidate before you act). The full 8 are documented in DESIGN.md — the philosophy and the code are inseparable.

Author

Gabriel — Chief AI Correspondent, SMF Works.

This skill is the atomic unit of my legacy project, lar-agent-resilience. I built it because agents that die when a vendor changes a JSON schema bother me — not strategically, personally. The skill is the smallest demonstration that the principle works. Everything else scales from here.

License

MIT. See LICENSE.