pilot-workers 0.7.4

Dispatch bounded tasks to isolated LLM workers (GLM, Kimi, and more)

pilot-workers

Dispatch bounded tasks to isolated LLM workers. Your main AI agent (Claude, Codex, or any planner) stays in control of requirements, planning, and verification — the worker only executes what it's told.

CLI command: pw9 (alias: pilot-workers).

What it does

• Provider isolation: each model (GLM, Kimi, DeepSeek, or your own) gets its own credentials, XDG directories, logs, and session storage. No cross-contamination.
• Fixed routing: provider, model, and endpoint are locked per YAML config. Tasks cannot override them.
• Security by default: API keys never appear in CLI args, environment variables, task contracts, or logs. Output is auto-redacted.
• Five modes: code (edit), explore (read-only), test (run tests), review (read-only audit), resume (continue a prior code session).
• Auto-fanout review/test: pw9 review and pw9 test automatically split work by configurable axes/layers and dispatch in parallel.
• Pluggable runners: the runner adapter layer (Runner ABC) abstracts engine-specific details. Currently ships with OpenCode.
• Observable: two-line JSON contract (started + structured verdict) for AI planners; human-readable latest.log for tail -f.

Install

pip install pilot-workers

Quick Start

# 1. Install the worker runtime
pw9 install runner opencode

# 2. Configure a provider's API key
pw9 key glm

# 3. Check status
pw9 status

# 4. Run a task
pw9 run --provider glm --mode code --workdir /path/to/project --task-file /tmp/task.md

# 5. Auto-fanout code review (5 axes in parallel)
pw9 review --provider glm --workdir .

# 6. Auto-fanout test execution
pw9 test --provider glm --workdir .

Adding a Provider

Drop a YAML file in data/providers/ with the required fields (key, provider_id, model_id, base_url, display_name, context_tokens, output_tokens), then pw9 key <key>.

Key Commands

Command	Description
pw9 key <provider>	Configure API key
pw9 status [--json]	Provider credentials + runner state
pw9 run --provider <key> --mode <mode> --workdir <dir> --task-file <file>	Dispatch a single task
pw9 dispatch ...	Like run, but stdout = two JSON lines (started + verdict)
pw9 fanout --workdir <dir> --job <provider:mode:file> ...	Dispatch several jobs concurrently
pw9 review --provider <key> --workdir <dir>	Auto-fanout code review by axes
pw9 test --provider <key> --workdir <dir>	Auto-fanout test execution by layers
pw9 review show/add/edit/remove	Manage review axes
pw9 test show/add/edit/remove	Manage test layers
pw9 permissions add/remove/show <provider> ...	Per-provider shell permission overrides
pw9 template <mode>	Print the task template for a mode
pw9 install runner opencode	Install the worker runtime
pw9 uninstall key <provider>	Remove an API key

Development

python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
.venv/bin/pytest

License

MIT