agentleFS
rumoiirumoii/rumo-engineering-skills/skills/rumo-clarify/SKILL.md

rumo-clarify

Skill7 starsChanged 14 days agoMITView on GitHub

Use before implementing a request whose goal, scope, acceptance criteria, or a risky decision is still unclear, such as a vague feature ask, an open-ended change, or a task touching data, permissions, deletion, or external publication without a stated target. Group the questions that can be answered now, with recommendations; follow up only for newly exposed consequential decisions, or explore deeper when the user asks. Do not use when the request is already specific, for pure questions, or for reviews.

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the goal, scope, and acceptance criteria are already clear, restate the understanding in one sentence and start.

    line 12
  2. Distinctive:

    Ask only questions that are answerable now.

    line 22
  3. Distinctive:

    When you end the turn with questions unanswered, end on the questions, not on a completion report.

    line 25
  4. Distinctive:

    When the user asks to dig in, go deeper, or keep asking, switch to rounds: each round asks only the questions whose prerequisites are now …

    line 36
  5. Very specific:

    Write an understanding summary of at most ten lines: goal, in scope and out of scope, acceptance criteria, assumptions, and blockers.

    line 40

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Rumo Clarify

Reach shared understanding with the fewest questions that matter, then get out of the way.

1. Decide Whether To Ask

If the goal, scope, and acceptance criteria are already clear, restate the understanding in one sentence and start. Asking about a clear request is a failure, not caution.

2. Look Before Asking

Answer facts yourself from the code, documentation, configuration, history, or environment. Ask the user only for decisions: intent, priorities, trade-offs, and authority.

3. Group Answerable Questions

  • At most five numbered questions, ordered by impact.
  • Each question carries a recommended answer and one line on what changes if the user picks differently.
  • Ask only questions that are answerable now; leave questions that depend on earlier answers for later.
  • In a plain-text round, tell the user they can reply "ok" to accept every recommendation listed in this round, or answer selectively, for example "2: use X". When the client offers a structured question tool, put the recommended option first and leave out the "ok" line.
  • "ok" accepts only the recommendations shown in this round. It never approves anything that was not asked.
  • When you end the turn with questions unanswered, end on the questions, not on a completion report.

4. Sort What Remains Open

  • Preference (naming, layout, wording, minor behavior): adopt the recommended default and list it as an explicit assumption.
  • Critical (target environment or database, deletion scope, permissions, irreversible operations, external publication, money or quota): do not default it. Pause the work that depends on it, say what is blocked, and keep going only on parts that do not depend on it.

Follow up only when an earlier answer exposes a consequential decision that cannot be resolved safely from the repository. Group the newly answerable decisions and explain what the answer changes. Do not repeat settled questions. When an answer changes the goal, restate the new goal first. Do not ask one decision per round as each comes up.

5. Dig Deeper Only On Request

When the user asks to dig in, go deeper, or keep asking, switch to rounds: each round asks only the questions whose prerequisites are now settled, in the same format. Stop when nothing is open or the user says enough.

6. Hand Off

Write an understanding summary of at most ten lines: goal, in scope and out of scope, acceptance criteria, assumptions, and blockers. Then implement directly for a small task, or move to rumo-plan for multi-step or cross-module work.

Example

Request: "Add an export feature to the order list."

Good round, after reading the code:

Understanding: an export button on the order list page.
Found: the customer list already exports CSV through the shared export helper.

1. Format? Recommended: CSV, like the customer list. Excel would add a dependency.
2. Which rows? Recommended: the current filter result. All orders would need a background job.
3. Who may export? Recommended: anyone who can view the list. Restricting it adds a permission.

Reply "ok" to accept all three, or answer by number, e.g. "2: all orders".

Weak round: "Which framework is the frontend? Where is the order list? CSV or Excel or both? What file name? Which columns? Which date format? ..." It asks facts the code answers, offers no recommendations, and buries the one decision that changes the design.

Authority

Answers to clarifying questions do not widen what the agent may do. Authorization the user already gave explicitly is not asked again.

Shortcuts And Rebuttals

Tempting shortcutWhy it fails
"I'll ask everything to be safe."Long questionnaires get skimmed; ask the five that change the outcome.
"I'll ask where the config lives."That is a fact you can look up.
"No answer on the target database, so I'll use the obvious one."Critical decisions are never defaulted; mark them blocked.
"They said ok, so I can also push and deploy.""ok" covers only the listed recommendations.
"The request is a bit vague, so I'll run the full interview."Ask the decisions needed now; an open-ended interview is the user's call.
"I'll ask about each decision as it comes up."Drip-fed rounds interrupt the user again and again; collect the open decisions and ask them together.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.