GitHub radar
Make your AI agent produce the same result every single run
A ready-made skill for your AI agent: it splits a complex task into steps controlled by code, not just prompts — each step hands off to the next on its own.
A tool that makes multi-step agent work predictable: plain code decides the order of steps, and the agent only works inside its own step.
When it helps
Useful if you run an agent through several stages (plan → build → review) and want the same result every time. Skip it for one simple request.
Pros
- Every step and event is logged — you can see what broke
- Model, prompts, and tools for each step live in one file
- Re-running gives the same result, not a random one
Cons
- Needs three separate API keys by default — OpenRouter, Fireworks, OpenAI
- Agents run through a separate program called pi, which you must install too
- Cutting it down to one key is possible, but takes manual setup
How to set it up — step by step
- 1Open your AI agent (Claude Code) inside your project folder.
- 2Send the agent the text from the block below — it will install the sssf skill and set up the process for your task.
- 3Copy .env.sample to .env and fill in the OPENROUTER_API_KEY key.
- 4Run `pi --version` to check the pi program (it runs the agents) is found — if not, point to it in .env.
- 5Ask the agent to run the test pass with `just demo` and check the log with `just obs`.
- 6Once the test run works, edit the agent prompts in adws/adw_data/prompt_engineering for your own task.
▌ More finds