Cyberwave Agent Skill

A portable Agent Skill for building and operating Physical AI systems on Cyberwave. One cyberwave entrypoint routes agents to focused guidance for:

• registration, authentication, and Cyberwave MCP connection
• environment creation and editing
• workflow authoring and run management
• safe robot control in simulation and live mode
• edge pairing, drivers, workers, and MQTT/Zenoh configuration
• asset onboarding and driver development
• robot telemetry, camera, workflow, and edge monitoring

The core follows the open Agent Skills format. Cyberwave MCP is the preferred typed execution plane when available, but the skill degrades to the verified SDK, CLI, dashboard, or official documentation.

Install

Canonical public skill (Claude Code or Codex)

git clone https://github.com/cyberwave-os/cyberwave-skill ~/.claude/skills/cyberwave
# Codex: clone to ~/.codex/skills/cyberwave instead

Use /cyberwave in Claude Code, $cyberwave in Codex, or describe a matching Cyberwave task and allow automatic discovery.

Project-local or other Agent Skills clients

Clone/copy the repository as a directory named cyberwave beneath the client's project or user Agent Skills search path. The required entrypoint is SKILL.md; provider-specific metadata is additive.

For ChatGPT or API runtimes that accept uploaded skill bundles, upload the same directory/ZIP and promote an immutable tested version. Keep MCP credentials in runtime configuration, not the bundle.

Cyberwave MCP

Hosted endpoint: https://mcp.cyberwave.com/mcp (Streamable HTTP). It uses a user-scoped Cyberwave API key in the Authorization: Bearer ... header. Configure the key with the client's secret mechanism; never commit it.

The skill discovers the tools actually exposed by the current client. MCP is optional for guidance and code authoring, but live platform execution and verification require an authenticated execution plane.

Architecture

SKILL.md is a compact orchestrator. It loads only the relevant module from references/ for the current task. This keeps the discovery and activation context small while retaining detailed domain procedures.

Validate

Run the portable validation before opening a pull request:

python3 scripts/validate_skill.py

To compare two skill checkouts before distribution:

python3 scripts/validate_skill.py --compare /path/to/cyberwave-skill

The evaluator scenarios in evals/scenarios.json define the expected routing and safety decisions for realistic user requests. Tests should assert those decisions and observable effects rather than exact prose.

License

Apache-2.0.