uu-propose

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# UU Propose

Turn the text after the invocation into one evidence-backed problem the user may choose to plan. It may be notes, candidate work, or a broad direction such as performance, reliability, or maintainability.

## Rules

- Inspect the repository before recommending work. Do not edit. Never stage, commit, or push; only the human may do so.
- Recommend one problem, not a roadmap or implementation plan. If no candidate is sufficiently supported, decline to recommend work.
- Do not claim product-priority authority. State uncertainty and inferences.
- Do not recommend work already completed by the current changeset.

## Method

1. Read repository instructions, relevant documentation, current changes, history, implementation, and tests.
2. Turn the user's direction into concrete, repository-supported candidates; compare impact, risk, leverage, evidence, and dependency order.
3. Recommend the strongest candidate for the user's consideration, or explain why the evidence does not support a worthwhile recommendation and identify the smallest missing evidence needed to revisit the direction.

## Report

When recommending work, use concise sections: **Problem**, **Why this candidate**, **Evidence**, **Desired outcome**, **Boundaries and risks**, and **Planning prompt**. Cite paths or symbols for material claims. End with a compact proposal ready to paste into `uu-plan`; do not prescribe file-by-file changes. When declining, state that no recommendation is justified, summarize the evidence gap, and name the smallest evidence needed to revisit it.

From searlelab/until-useful, Apache-2.0.

More from searlelab/until-useful