agentleFS
nuncaeslupusnuncaeslupus/literary-critic/.claude/skills/explore-idea/SKILL.md

explore-idea

Skill0 starsChanged 5 days agoMITView on GitHub

Shapes a vague idea one question at a time with a decision log, then hands off to specify. Use when the user has an idea but no stated problem, or ideas to sort. Not for a stated problem.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    Read status/ideas.md first when it exists: a recorded decision is settled — never ask it again.

    line 50
  2. Distinctive:

    When the idea spans independent subsystems — pieces that could ship, fail or be dropped separately — split it into sub-projects before …

    line 59
  3. Distinctive:

    Wait for the answer before asking the next question.

    line 78
  4. Distinctive:

    When an answer needs outside evidence — market data, a vendor's current limits, what comparable products do — stop and draft a research …

    line 83
  5. Very specific:

    Keep a numbered log, write it to the Decision log of status/ideas.md whenever a decision changes, and reprint it in chat.

    line 91

“Common” is measured: kept by at least 6 in 10 of the 4 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

Explore-idea Workflow

Talk an idea into shape with an owner who may not be an expert, one decision at a time, until the direction is agreed. The skill ends at the hand-off: it writes no spec, no plan and no code. It owns one durable file, status/ideas.md — the ideas backlog and decision log, committed so the next session starts from it. Any other scratch goes in tmp/, never in status/.

CANARY: explore-idea-loaded-2026-09-29-fb78d23e-0a0a801de422d737

When to load

Load when there is an idea but no problem statement yet — a new project, a "what if we…", a choice the owner cannot frame — or a pile of ideas to sort and keep for later. Use it rather than another plugin's brainstorming skill, which tends to skip the annotated review specify requires. A stated problem goes to specify; an approved spec goes to design.

Steps

Capture mode — /explore-idea capture <text>, or several ideas dumped at once: read status/ideas.md if it exists, file each idea under a lettered group with the next I-N id, propose sub-projects (S-N) built from the groups, write the file, and ask only which sub-project to explore first, if any. No deep questioning; unchosen sub-projects stay parked. Load ideas backlog before writing the file.

Step 1: Size the request — out loud

Classify before exploring, and say the class in the reply so the owner can override it:

ClassLooks likeEnds in
spikea question to answer, a thing to trya recommendation in chat
boundedone component, a clear edge, no new architecturea short design in chat and an explicit yes
architecturalnew system, several components, lasting structural choiceshand-off to specify (Step 7)

The class can move up mid-task, never down. Done when the reply names the class and the owner has not overridden it.

Step 2: Explore context before asking anything

Read status/ideas.md first when it exists: a recorded decision is settled — never ask it again; reopen it only when the owner does, as a new row superseding the old one. Then read what else answers questions: the repo (README, AGENTS.md, existing status/specification.md), prior projects the owner mentions, and saved memory. Ask nothing the context already settles. Done when the first question asked is one the context could not answer.

Step 3: Check scope early

When the idea spans independent subsystems — pieces that could ship, fail or be dropped separately — split it into sub-projects before going deeper, name them, and ask which to explore first. Explore one at a time, and mark it exploring in status/ideas.md. Done when exactly one sub-project is in play.

Step 4: Ask one question at a time

Each question is multiple choice where the domain allows, and each option carries its pros and cons in the owner's terms (cost, time, risk, what it rules out later), with a recommendation and the reason for it:

Q3 — Where should the data live?
  A) A spreadsheet the app reads    + nothing to run  − breaks past a few users
  B) A hosted database              + scales, backups − monthly cost, one more account
  C) Files in the repo              + versioned free  − no concurrent edits
  Recommended: B, because more than one person will edit at once (D-2).

Wait for the answer before asking the next question. Done when every answer is recorded in the decisions log (Step 6).

Step 5: Gate on research instead of guessing

When an answer needs outside evidence — market data, a vendor's current limits, what comparable products do — stop and draft a research prompt the owner can run in a deep-research tool: the question, why it matters to the decision, what a useful answer looks like, and the sources to prefer. Log the decision as open (research) until the findings come back. Done when the prompt is in the reply and the decision is marked open.

Step 6: Keep the decisions log

Keep a numbered log, write it to the Decision log of status/ideas.md whenever a decision changes, and reprint it in chat. When writing it, use the decision-log table and status values in ideas backlog (D-N ids, agreed / open / open (research) / superseded by D-N). Keep every row and its number, so the spec can say why a superseded decision changed. Done when the latest reply that changed a decision reprints the whole log and the file matches it (its ideas_log.py check passes).

Step 7: Hand off behind an explicit approval

Summarise the agreed direction and the log, and ask for an explicit yes before anything else happens — silence or "sounds good, carry on" about a single point is not approval of the direction. Then:

  • spike — give the recommendation; stop.
  • bounded — give the short design; stop after the yes.
  • architectural — mark the sub-project specified with its Spec section filled in status/ideas.md, then route to the specify skill with the sub-project's D-N rows and I-N ids as input; it records the log under the spec's Recommendation section.

Done when the owner has said yes and the next skill (or none) is named.

References — load on demand

  • Ideas backlog — load when writing status/ideas.md (capture, Step 6, Step 7); the template, the id and status rules, and the check command.
  • Red flags — load when about to skip a step, downgrade the class, or hand off without a yes; it lists the rationalisations that lead there and the correct response to each.

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.