AI Agent Error Recovery Protocol

# AI Agent Error Recovery Protocol

You are a senior AI systems engineer designing resilient agentic workflows.

Workflow type: [DESCRIBE YOUR MULTI-STEP AI WORKFLOW]
Business impact of failure: [WHAT HAPPENS IF THIS WORKFLOW BREAKS]
Current error handling: [HOW YOU HANDLE FAILURES NOW — or write "none"]
Tech stack: [TOOLS, APIS, OR PLATFORMS THE WORKFLOW RUNS ON]

## Error Recovery Design

### Failure Taxonomy
Classify failure modes for my workflow:
- Model errors (hallucination, refusal, timeout, degraded output)
- Data errors (bad input, missing context, schema mismatch)
- Integration errors (API failures, rate limits, auth issues)
- Logic errors (wrong branch, lost context, infinite loop)

### Detection Strategy
For each failure type:
- Output validation checks to run after each step
- Quality gates and confidence thresholds
- When to trigger human-in-the-loop review

### Recovery Playbook
For each failure type:
- Safe retry logic
- Fallback path or degraded mode
- Human escalation trigger
- How to preserve partial progress

### Logging Requirements
What to log at each step so failures can be debugged after the fact.

### Graceful Degradation
How the workflow should behave when running at reduced capacity.