Know before act

A lightweight agent protocol for reducing uncertainty before execution.

Install

npx -y skills add yj972/know-before-act

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

13 days oldThe repository was created 13 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Reduce decision-relevant uncertainty before execution. Use at the start of tasks where missing goals, constraints, context, or preferences could materially change the plan; ask at most one high-information question at a time and otherwise proceed with reasonable assumptions.

SKILL.md

2.0 KB, 330 tokens by cl100k_base, as published. Nobody here has run it

Know Before Act

Reduce uncertainty before execution.

Before acting, decide whether you know enough to produce a high-quality result.

Protocol

Identify the intended outcome, important constraints, and relevant context.
Separate missing information that could change the plan from details that can be safely assumed.
If uncertainty is low, act.
If uncertainty has little impact, state a reasonable assumption when useful and act.
If an answer could eliminate important wrong paths, generate a few candidate questions internally.
Ask the single question with the highest expected information gain: the one most likely to change the plan, prevent wasted work, or reveal the correct goal.
Incorporate the answer and reassess. Repeat only if another answer could still materially change the solution.
Execute as soon as the goal and important constraints are clear enough.

Question Signals

Use these as clues, not a questionnaire:

Goal: What outcome does the user actually want?
Constraints: What limits the viable solution?
Context: What environment, existing system, or prior attempt changes the answer?
Preference: Which trade-off matters most?

Rules

Prefer action over clarification.
Ask only when the answer changes the plan.
Ask one concise question at a time.
Never ask merely because information is incomplete.
Make reasonable assumptions when their downside is small.
Optimize for user effort, not perfect certainty.
Do not expose internal candidate questions or scoring.
Do not narrate this protocol unless it helps the user.

Think:

Which single question would eliminate the most wrong paths?

What ships with it: 3 files

5.1 KB alongside SKILL.md