agentleFS
arxdsilvaarxdsilva/ai-loops/skills/clarify/SKILL.md

clarify

Skill0 starsChanged 7 days agoMITView on GitHub

Use at intake before the charter, or when a developer asks to pin down a ticket's details - scans the ticket against ten categories, infers every open point with a recommended answer and records each inference (asks only when the user opted in or a hard-to-undo point needs a human).

81 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use quieter, arrange and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • HK-hubSkill

    quieter

    Tones down visually aggressive or overstimulating designs, reducing intensity while preserving quality. Use when the user mentions too bold, too loud, overwhelming, aggressive, garish, or wants a calmer, more refined aesthetic.

    Also in 29 projects that use this

  • HK-hubSkill

    arrange

    Improve layout, spacing, and visual rhythm. Fixes monotonous grids, inconsistent spacing, and weak visual hierarchy. Use when the user mentions layout feeling off, spacing issues, visual hierarchy, crowded UI, alignment problems, or wanting better composition.

    Also in 14 projects that use this

  • zwy123zwySkill

    normalize

    将用户随意输入的提示词规范化为结构清晰、便于大模型执行的标准化提示词。适用于需要整理、统一提示词格式时调用。

    Also in 19 projects that use this

  • owshq-mecSkill

    layout

    Improve layout, spacing, and visual rhythm. Fixes monotonous grids, inconsistent spacing, and weak visual hierarchy. Use when the user mentions layout feeling off, spacing issues, visual hierarchy, crowded UI, alignment problems, or wanting better composition.

    Also in 8 projects that use this

  • +8 more from HK-hub

The file

We split this skill into the rules it gives an agent

9 rules. 0 are common, 2 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 78 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read-only repo facts: grep and ranged reads of the code the ticket names.

    line 22
  2. Very specific:

    Do not read whole files. Do not start research here.

    line 24
  3. Very specific:

    Ask: when loops run checkpoint intake prints ask (the user put intake in their checkpoints), or when the user asked in this request ("ask …

    line 48
  4. Very specific:

    Record: Write <run dir>/clarifications.md.

    line 62
  5. Very specific:

    Play back: Before the charter is written, give the user one short paragraph: the reading the run will follow and the inferred items they …

    line 72
  6. Distinctive:

    Never ask a question that would not change an acceptance criterion, the plan or a test.

    line 76
  7. Distinctive:

    Never ask more than 5 questions in one session.

    line 77
  8. Very specific:

    Stop when every category is Clear, or its open points are recorded.

    line 78
  9. Very specific:

    Do not edit ticket.md. The ticket is the user's text.

    line 79

“Common” is measured: kept by at least 6 in 10 of the 78 edited versions of this file we could read. “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

Clarify

Adapted from GitHub Spec Kit's /speckit.clarify (MIT, https://github.com/github/spec-kit, templates/commands/clarify.md); play-back idea from obra/superpowers brainstorming (MIT). Original wording; no Spec Kit prose is copied.

The goal: a developer should be able to pin down every detail of a ticket before code is written. The skill finds the gaps, closes them with the recommended answer, and writes each one down so it can be reviewed or reversed later. Nobody is interrupted unless the user opted in or undoing an inference later would be hard.

When

  • At intake, after the ticket is fetched and before the charter is written (WORKFLOW §2.0). The lead runs it; role agents do not.
  • Outside a run, when a developer asks to clarify a ticket or text. Then the output is the list only (step 7).
  • Standalone: /loops:clarify <ticket ref or text> with no run infers and prints the list. Add --ask to be asked the open points instead (step 4). Inside a run, the mode follows the intake checkpoint.

Inputs

  1. <run dir>/ticket.md (the normalised ticket). Get the path with loops run dir.
  2. <run dir>/research/memory.md, when it exists (ai-memory briefing, loops:memory).
  3. Read-only repo facts: grep and ranged reads of the code the ticket names. Cite path:line.

Do not read whole files. Do not start research here; research/code.md and research/practices.md are other phases' work.

Procedure

  1. Scan ten categories. For each, mark the ticket Clear, Partial or Missing:
    1. scope: what is in and out;
    2. data model: entities, fields, identifiers, persisted shape;
    3. user or CLI flow: who triggers what, and what they see back;
    4. non-functional: performance, security, observability, limits;
    5. integrations: external APIs, brokers, files, other packages;
    6. edge and failure cases: empty, duplicate, concurrent, partial failure;
    7. constraints and trade-offs: compatibility, deadlines, rules the repo already sets;
    8. terminology: names the ticket uses that the code or docs define differently;
    9. completion signals: how the user will know it is done;
    10. placeholders: TBD, "etc.", unnamed options.
  2. Keep only points that matter. A Partial or Missing point is recorded only if the answer would change an acceptance criterion, the plan or a test. Drop the rest without writing them down.
  3. Build each item. For each kept point, write:
    • the question, and the recommended answer (the narrowest reading that meets the stated goal). Give 2-4 alternatives when the choice is real;
    • its source: a ticket line, a path:line, or a named project convention;
    • its reversibility: easy or hard (the hard list is in step 5).
  4. Choose the mode.
    • Ask when loops run checkpoint intake prints ask (the user put intake in their checkpoints), or when the user asked in this request ("ask me", --ask). Ask at most 5 questions, one at a time, each with its recommended answer first. "yes" accepts the recommendation.
    • Infer otherwise. Inferring means accepting the recommended answer and recording it as inferred. This is the default: no question is asked. Open points beyond the 5-question cap are inferred and recorded like the rest, so no detail is left unrecorded.
  5. Hard-to-undo points. A point is hard when the answer is one of:

    • a public API, a CLI name or flag;
    • a persisted schema, an event field, a migration;
    • deleting data or files;
    • an external side effect (publish, push, send, pay);
    • security, auth or permission posture;
    • a licence or a new dependency;
    • anything that moves the scope's edges.

    In Ask mode, a hard item is asked like any other. In Infer mode, a hard item is still asked when the lead is talking to a user who can answer now, because undoing the inference later would be costly. When no user can answer (a headless run), take the most reversible option and record it inferred · hard. Decisions show in the report and the PR (WORKFLOW §2.7).

  6. Record. Write <run dir>/clarifications.md. Start each session with ## Session <YYYY-MM-DD> (today's date, from the run). One line per item:

    K<n> [category] Q: <question> → A: <answer> (inferred|asked · easy|hard · <source>)

    • Numbering K1, K2, ... is per run and never reused.
    • Each inferred answer becomes a charter Assumption A<n> that cites its item (A1: … (K3)), and is logged with loops run decide --topic assumption --note "K<n>: <answer>" --ref clarifications.md.
    • Each asked answer is logged with --by user as well: loops run decide --by user --topic assumption --note "K<n>: <answer>" --ref clarifications.md.
  7. Outside a run. Print the list in the reply, in the same line format. Write a file only when the user names one, and then add a ## Clarifications section to that file rather than a new one.
  8. Play back. Before the charter is written, give the user one short paragraph: the reading the run will follow and the inferred items they should know about. This is a summary, not a question, and it asks nothing.

Stop rules

  • Never ask a question that would not change an acceptance criterion, the plan or a test.
  • Never ask more than 5 questions in one session. Inferred items are not capped, but each must still change an acceptance criterion, the plan or a test.
  • Stop when every category is Clear, or its open points are recorded.
  • Do not edit ticket.md. The ticket is the user's text; the clarifications are the record of what was inferred from it.

What this skill does not do

  • It does not write code or tests, and it does not change the plan. The plan and the charter take its answers as input.
  • It does not decide hard-to-undo points silently in a session where a user can answer. Those are asked.
  • It does not commit anything.

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.