Construct-AI-primary Integration - Cross-Agent Compatibility Testing
Overview
Integration tests cross-agent compatibility for PromptForge AI. A prompt that works perfectly for one agent might fail for another. Integration ensures prompts are compatible across all agents, models, and company boundaries that will use them.
When to Use
- Before deploying a prompt that will be used by multiple agents
- When testing prompt compatibility across different AI models
- When validating cross-company prompt interoperability (PromptForge ↔ DevForge ↔ QualityForge)
- When adding a new agent to the ecosystem
- When troubleshooting compatibility issues between agents
- Don't use when: Testing a prompt used by a single agent in isolation
Core Procedures
Step 1: Identify Compatibility Matrix
Map out what needs to be tested:
- Agents: Which agents will use this prompt?
- Models: Which AI models will run this prompt? (Claude, GPT, Gemini, etc.)
- Companies: Which companies will share this prompt?
- Versions: Which versions of each agent/model need testing?
Step 2: Design Compatibility Tests
For each cell in the compatibility matrix:
COMPATIBILITY TEST
==================
Prompt: [name and version]
Agent: [agent name and version]
Model: [model name and version]
Company: [company context]
Test Inputs: [representative inputs for this agent/model]
Expected Output: [what this agent/model should produce]
Acceptance Criteria: [what "compatible" means for this test]
Step 3: Execute Compatibility Tests
Run the prompt against each agent/model combination:
- Send identical prompt to each agent/model
- Record outputs and any errors
- Compare outputs against acceptance criteria
- Note any model-specific or agent-specific behavior differences
- Document compatibility status for each combination
Step 4: Resolve Compatibility Issues
For each incompatibility found:
- Minor: Output differs slightly but is acceptable → Document the difference, accept
- Moderate: Output quality varies significantly → Optimize prompt for the weaker model/agent
- Severe: Prompt fails entirely on specific model/agent → Create model-specific variant or fix root cause
Step 5: Report Compatibility Status
COMPATIBILITY REPORT
====================
Prompt: [name and version]
Test Date: [date]
Matrix Size: [agents × models × companies = total tests]
Passed: [count]
Failed: [count]
Warnings: [count - minor differences]
Overall Status: [fully compatible / partially compatible / incompatible]
Incompatible Combinations: [list with details]
Recommended Actions: [what to fix and priority]
Success Metrics
- Compatibility coverage: 100% of shared prompts tested across all agent/model combinations
- Compatibility pass rate: >=95% of tests pass on first run
- Resolution time: <24 hours to resolve critical incompatibilities
- Cross-company compatibility: 100% of shared prompts work across company boundaries
Error Handling
- Error: Prompt works on Claude but fails on GPT Response: Analyze the difference (model capability vs. prompt structure), create model-specific variant if needed, or redesign prompt to be model-agnostic
- Error: Agent-specific context causes different behavior Response: Add agent-specific instructions or context handling, or standardize the context format across agents
- Error: Cross-company handoff changes prompt behavior Response: Test the handoff protocol, ensure context transfer preserves all necessary information
Cross-Team Integration
Related Skills: flowdesigner-promptforge-agent-handoff, harmonic-promptforge-orchestration-strategy, collaborator-promptforge-cross-company-knowledge-sharing, validator-promptforge-syntax-logic-validation Used By: FlowDesigner, Harmonic, Collaborator, Validator, ALL agents sharing prompts across boundaries