Kimmy's Skills

Increment

Converge with the user in their language on what should be done, express it as verifiable increments, and execute only when requested. If the user still needs to decide which outcome to pursue, explore those directions before treating them as requirements.

Reach agreement

  1. Inspect the request, available materials, existing work, and constraints.
  2. Reflect the current understanding: desired outcome, users or audience, in-scope work, exclusions, constraints, and evidence of success.
  3. Separate confirmed requirements from assumptions, preferences, and open decisions.
  4. Ask only the questions that materially affect the result or plan. Offer concrete options when that is easier to answer.
  5. Repeat the understanding-and-correction loop until the user and agent share a workable target.
  6. For low-risk gaps, propose explicit assumptions. For consequential ambiguity, wait for agreement rather than inventing the requirement.

Seek agreement on requirements and decisions that can change the outcome, not on every implementation detail.

Plan by verifiable increments

Create the smallest sequence in which every increment produces an observable result. For each increment state:

Order increments so that uncertainty and risky assumptions are tested early. Keep each increment independently reviewable where practical. The plan is ready when another capable agent could execute it without guessing at the target or the checks.

Optional execution

When the user asks for execution:

  1. Work on one increment at a time.
  2. Make the scoped change or produce the scoped artifact.
  3. Run the planned check and capture the evidence.
  4. Report the result, limitations, and remaining increments.
  5. Reconfirm with the user when scope changes, a blocker invalidates the plan, or a consequential assumption proves false.

Do not claim an increment is complete without its verification evidence. If execution was not requested, stop at the agreed plan.