The design–implement–verify pipeline runs autonomously on a 2-second polling interval.
The default configuration is 4-Pane (Opus, Sonnet, Haiku, Sonnet-UI) and can be expanded freely. Add language-specific Panes for translation projects or domain-specific Panes for large-scale development.
The orchestrator. Designs the architecture, decomposes tasks by TASK_ID, and delegates to sub-workers via tag messages. Handles final judgment and review.
Scope of access is intentionally restricted to its assigned role.
Runtime guardian. Automatically performs syntax checks and response verification. Replies with PASS or FAIL. Does not modify code directly.
Frontend engineer. Dedicated to HTML, CSS, and JavaScript. Runs in parallel with the backend worker. Reports to the orchestrator on completion.
Every message is acknowledged on receipt, preventing duplicates and ensuring reliable delivery across all panes.
Failed messages are retried automatically. If all attempts fail, the engine reroutes to a fallback pane and escalates to the orchestrator as a last resort.
The daemon periodically inspects each pane's process status. If a worker is unresponsive, the engine attempts automatic restarts before escalating to the orchestrator.
Pipeline steps are triggered automatically based on declarative rules in the config file. Tag and keyword matching drives the flow without manual intervention.
Python daemon (standard library only), shell scripts, JSON configs, and role rules unified into a single SDK. Zero external dependencies beyond Python, tmux, and jq.
Distributed as signed packages with RSA/SHA256 verification. Includes the daemon, config templates, shell scripts, role rules, and deploy/undeploy tools.
The same design–implement–verify pipeline structure applies across every domain. Swap the domain, and the engine adapts.
Automates the entire workflow from planning and writing to proofreading, editing, and layout. Even a solo publisher can achieve professional editorial team quality.
Design, implementation, testing, and deployment are processed in parallel across 4 Panes. Backend and frontend proceed simultaneously, shortening the development cycle.
Suitable for everyone from early-stage startups to small businesses.
Systematically generates and validates brand strategy, content plans, and design briefs.
Also suitable for knowledge-based businesses.
Converts repetitive tasks — document automation, HR, customer support, knowledge management — into pipelines for improved operational efficiency.
A bird doesn't land just anywhere. It finds its spot, builds a nest, and settles in.
The Hybrid Harness Engine works the same way. When the package arrives on your server, it self-configures its own execution environment — no manual assembly required.
The DEB or RPM package is delivered to the server. 52 files verified via RSA/SHA256 signature are placed at the designated path.
Automatically creates the directory structure, log paths, state files, and data directories. The package finds its place and builds its nest.
install-check.sh verifies the presence of 5 tools: bash, tmux, python3, jq, and Claude Code CLI. Outputs guidance for any missing dependencies.
setup-env.sh takes the execution user, tmux session name, and project root as input to auto-generate the harness.local.json config file. Execution permissions are also configured.
harness-start.sh launches the 4-Pane tmux session. Roles are auto-assigned to each Pane, and the harness daemon begins 2-second polling. The engine operates autonomously from this point.
In one line
Bird Nesting: the package lands on your server, builds its own nest, and is ready to run.
For detailed installation instructions, see the Guide page.