Engines are the workers in the Cabin Crew platform—specialized binaries that execute specific tasks (code generation, infrastructure planning, testing) within the governed workflow.

Engines are distributed as OCI images (Docker format) but executed as subprocesses on the host:

1. Orchestrator pulls the image from GHCR/Docker Hub
2. Extracts the static binary (FROM scratch)
3. Runs the binary directly on the host OS

This "extraction strategy" allows engines to share host tools (git, npm, terraform) without containerization overhead.

All engines communicate via a standardized JSON protocol:

• Input: Issue metadata, configuration, environment variables (via stdin)
• Output: Artifacts, logs, exit codes (via stdout)

The protocol is versioned and documented in the cabincrew-protocol repository. Any tool that speaks this protocol can become an engine.

Cabin Crew maintains a core set of open-source engines (Apache 2.0):

• Scout Engine: Scans file tree and tech stack, generates deterministic context
• Planner Engine: Parses GitHub Issues, negotiates requirements, manages DAG via HTML comments
• Dev Engine: Wraps Aider, runs TDD loop (Write → Test → Fix), outputs patches
• Infra Engine: Wraps Terraform, splits Plan (artifact) from Apply (execution)
• Tower Engine: Wraps GitHub Approvals API for human gates ("tower-clearance")

You can wrap any script into the governance flow:

1. Write a wrapper that reads JSON from stdin
2. Execute your custom logic (bash, Python, whatever)
3. Write results as JSON to stdout
4. Package as an OCI image (FROM scratch)
5. Sign with Cosign

Example: A bash-engine that runs legacy deployment scripts within the governed workflow, gaining audit logs and policy enforcement without rewriting the scripts.