Codehornets is a local Python toolkit for compiling complete agent context packages from structured task files. v0.1 is the foundation: a CLI that takes a task envelope (JSON) and produces a single text bundle combining the harness system prompt, the assigned agent's role, the relevant project context, the task itself, and the expected output contract.

With uv:

uv venv
uv pip install -e ".[dev]"

Or with pip:

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Compile a task file to stdout:

codehornets compile-context tasks/examples/create_cli.json

Write the compiled context to a file:

codehornets compile-context tasks/examples/create_cli.json --output build/context.txt

Run against a different project root (one containing a contexts/ directory):

codehornets compile-context path/to/task.json --root /path/to/project

pytest

• TaskEnvelope and AgentResult Pydantic schemas with validation.
• A ContextBuilder that loads system, role, project, task, and output-contract sections in a fixed order.
• A Typer CLI exposing compile-context.
• Starter context content for builder, architect, and reviewer roles and an audacity-copilot example project.
• Pytest coverage for schemas and the context builder.

• No dashboard or web UI.
• No Redis, queues, or background workers.
• No Docker images.
• No vector or long-term memory.
• No multi-agent orchestration.
• No OpenAI / Anthropic API calls — this version only assembles the prompt context.