Interactive guide · working sketch
What shape is this task?
Describe the task and set six dials. The workflow contract updates as you move them: what stays human, what the agent runs, and what evidence you need before accepting the result.
Shape the task
Can you describe what “done” means before work starts?
Does success depend on taste, product judgment, or a decision only you can make?
What happens if the agent confidently gets this wrong?
Can the result be tested or compared against an objective signal?
Can you inspect, undo, or isolate the change before it affects anyone?
If this works, is it worth turning the path into a reusable system?
Contract detail
Useful next tool
Recommended resource
Preview the task brief
After you run it: was this enough ceremony?
Your answer is saved only in this browser. It helps you notice where the suggested workflow was too light or too heavy.
Let your agent shape the task for you
Works with any agent. The skill inspects your current context, suggests the six inputs, calls the deterministic API, and returns the workflow plus instructions for any recommended skills or plugins.
Your task description and repository context stay local. The endpoint only needs six numeric values and explicit hazard signals.
/task-workflow:cca-task-shape
Your recent task shapes
Nothing saved yet. The useful part is comparing tasks over time, not chasing one perfect score.
More autonomy requires clearer work, not longer prompts.
I do not trust agents with important work directly. I trust small, tested systems that make the agent see the right context, produce evidence, and stop at the decisions that are actually mine.
Use the browser when you want to think through the dials yourself. Use the skill when your agent already has the context and can do the legwork.