agentshift 0.4.0

CLI transpiler for converting AI agents between platforms

AgentShift

Convert AI agents between platforms. Define once, run anywhere.

---

Your OpenClaw skill shouldn't be locked to one platform. AgentShift converts it to Claude Code, GitHub Copilot, AWS Bedrock, Microsoft 365 Copilot, GCP Vertex AI, and more.

Install

# pip
pip install agentshift

# from source
git clone https://github.com/ogkranthi/agentshift.git
cd agentshift && pip install -e .

Quick start

# → Claude Code
agentshift convert ~/.openclaw/skills/weather --from openclaw --to claude-code --output ./weather-claude

# → GitHub Copilot
agentshift convert ~/.openclaw/skills/weather --from openclaw --to copilot --output ./weather-copilot

# → AWS Bedrock
agentshift convert ~/.openclaw/skills/weather --from openclaw --to bedrock --output ./weather-bedrock

# Convert to ALL supported targets at once
agentshift convert ./my-skill --from openclaw --to all --output ./output
# → output/claude-code/  output/copilot/  output/bedrock/  output/m365/  output/vertex/

# Validate generated output before deploying
agentshift validate ./output/bedrock --target bedrock

weather-claude/               weather-copilot/
├── CLAUDE.md                 ├── weather.agent.md
└── settings.json             └── README.md

How it works

1. Parse  →  SKILL.md · CLAUDE.md · .agent.md · AGENTS.md · bot-meta.xml · bedrock-agent.json
              ↓
2. IR     →  identity · tools · knowledge · triggers · constraints · governance
              ↓
3. Emit   →  Claude Code ✅ | Copilot ✅ | Bedrock ✅ | M365 ✅ | Vertex ✅ | A2A ✅ | LangGraph ✅

Parse cloud agent artifacts

# Parse Bedrock artifacts → convert to OpenClaw skill
agentshift convert ./bedrock-output/ --from bedrock --to openclaw --output ./my-skill

# Parse Vertex AI artifacts → convert to Claude Code
agentshift convert ./vertex-output/ --from vertex --to claude-code --output ./claude-output

# Parse Copilot .agent.md → convert to Bedrock
agentshift convert ./my-copilot/ --from copilot --to bedrock --output ./bedrock-output

# Diff portability from Bedrock source
agentshift diff ./bedrock-output/ --from bedrock --targets claude-code,copilot

# Governance audit: Vertex → Bedrock round-trip
agentshift audit ./vertex-output/ --from vertex --targets bedrock

# Generate A2A Agent Card for platform interoperability
agentshift convert ~/.openclaw/skills/weather --from openclaw --to a2a --output ./weather-a2a

New in v0.4.0

# EU AI Act compliance check
agentshift compliance ./my-agent --from claude-code --framework eu-ai-act

# Machine-readable portability scores
agentshift diff ./my-agent --from openclaw --output-format json

# Local agent registry
agentshift registry register ./my-agent --name "github-assistant" --from openclaw

Using with GitHub Actions

# Add to .github/workflows/portability.yml
- uses: ogkranthi/agentshift/.github/actions/portability-report@main

Automatically comments portability scores on PRs.

See portability before converting

agentshift diff ~/.openclaw/skills/github --from openclaw

Component          Source    claude-code      copilot        bedrock
─────────────────────────────────────────────────────────────────────
Instructions         ✅       ✅ 100%         ✅ 100%      ✅ 100%
Tools (shell: 2)     ✅    ✅ Bash(bin:*)   ✅ terminal   ⚠️  Lambda*
─────────────────────────────────────────────────────────────────────
Portability                    100%             92%           38%

Governance layer

AgentShift v0.3 introduces a three-layer governance model that travels with your agent through every conversion:

Layer	Model	Source
L1 — Guardrails	Guardrail — prompt-level safety rules	SOUL.md, instruction.txt, Bedrock topics, Vertex instructions
L2 — Tool permissions	ToolPermission — per-tool access control	OpenClaw tools/*.json
L3 — Platform annotations	PlatformAnnotation — native filters	governance/annotations.json, Bedrock guardrail config

Governance is preserved and audited during conversion:

# Audit governance preservation: bedrock → claude-code
agentshift audit ./my-bedrock-agent/ --from bedrock --targets claude-code

# Audit from Vertex AI artifacts
agentshift audit ./vertex-output/ --from vertex --targets bedrock,claude-code

Agent registry

Track your agents and detect configuration drift across environments:

# Register an agent
agentshift registry register ~/.openclaw/skills/weather

# List all registered agents
agentshift registry list

# Detect drift since last registration
agentshift registry diff weather

# Export registry as JSON
agentshift registry export

Registry is stored at ~/.agentshift/registry.json and works offline.

Supported platforms

Platform	Read (parser)	Write (emitter)	Status
OpenClaw	✅	✅	Works today
Claude Code	✅	✅	Works today
GitHub Copilot	✅	✅	Works today
AWS Bedrock	✅	✅	Works today
GCP Vertex AI	✅	✅	Works today
AGENTS.md	✅	—	Works today
Salesforce Agentforce	✅	—	Works today
Microsoft 365	—	✅	Works today
Google A2A	—	✅	Works today
LangGraph	—	✅	Works today

Guides

Target	Guide	Examples
Claude Code	docs/claude-code.md	examples/weather-to-claude-code
GitHub Copilot	docs/copilot.md	examples/github-to-copilot
AWS Bedrock	docs/bedrock.md	examples/github-to-bedrock
Microsoft 365	docs/m365.md	examples/github-to-m365
GCP Vertex AI	docs/vertex.md	examples/github-to-vertex
Architecture	docs/architecture.md	—

Contributing

Contributions welcome — especially new platform parsers/emitters.

See CONTRIBUTING.md for setup, architecture, and PR guidelines.

git clone https://github.com/ogkranthi/agentshift.git
cd agentshift && pip install -e ".[dev]"
agentshift --help

Open a Platform Request to discuss a new target.

License

Apache License 2.0