agentleFS
marinvchmarinvch/Cortex/skills/clarify-request/SKILL.md

clarify-request

Skill1 starsChanged 18 days agoMITView on GitHub

Sharpen a vague or short prompt ("fix it", "make it faster") before work starts, or when the prompt hook scores its vagueness 4 or higher. Triggers — "optimize this prompt", "what should I have asked".

2 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

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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:

    Never act on an unconfirmed prompt.

    line 69
  2. Distinctive:

    Don't ask more than 2 questions.

    line 104
  3. Distinctive:

    Don't fire on slash commands, confirmations, or mid-flow replies.

    line 105
  4. Distinctive:

    Don't act on a prompt the user hasn't confirmed.

    line 106
  5. Distinctive:

    Don't optimize a prompt that is already precise — silence is the correct output.

    line 107
  6. Very specific:

    Don't write the record anywhere but docs/prompts/ (gitignored.

    line 108

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

/clarify-request — sharpen a prompt before acting on it

A vague prompt makes the agent guess, and a guess costs a whole turn. This ritual converts an unclear ask into one precise prompt, gets it confirmed, records it, and hands it to the right ritual.

What to do

  1. Score it (the hook does this in Claude Code; do it yourself everywhere else): under 10 words +2; no action verb +1; no component reference +1; no domain keyword +1. This file is where these rules live — .claude/hooks/optimize-prompt.mjs implements them, and its test fails if a word below goes missing from this list.

    • Action verbs, in any inflection (merged, fixing): add, create, update, delete, fix, remove, migrate, refactor, write, build, review, audit, explain, document, test, rename, move, debug, optimise/optimize, install, scan, implement, generate, wire, split, merge, run, restore, resume, continue, investigate, research, interview, check, verify, compare, analyse/analyze, summarise/summarize, plan, ship, commit, push, deploy, clean, finish, list, find, search, show, open.
    • Component reference: a path, a `backticked` token, a file.ext, a #123, a URL, a /ritual anywhere in the prompt, or a hyphenated ritual name written bare (review-change). Single-word ritual names (ship, daily) don't count bare — they are ordinary English.
    • Domain keywords: auth, db, database, api, ui, schema, test(s), hook(s), skill(s), vault, graph, mcp, git, ci, cli, doc(s), readme, agent(s), prompt(s), session(s), pr(s), branch(es), commit(s), worktree(s), plugin(s), ritual(s), cortex, memory, index, eval(s), changelog, repo(s), release(s).

    Under 4 — act on the prompt as written, say nothing. Bypass entirely (no score, no directive) when the prompt:

    • is empty, starts with /, is over 2000 characters, or is over 60 words;
    • is a steer of two words or fewer (yes, ok, go ahead, stop, …), or the same typed in Cyrillic (ок, да, добре, давай, продължи);
    • is a word of approval alone, or in front of a go-ahead phrase: super, great, perfect, nice, cool, awesome, excellent (super do it, great, go ahead). In front of anything else (great work on the graph) the prompt still scores;
    • is a go-ahead of eight words or fewer — go ahead, carry on, sounds good, agreed, lgtm, yes please, ok/yes/sure + any action verb above (or do/go/proceed), or do it/all/them/both/the rest — followed by anything (go ahead do all of them, ok merge it, yes write the spec). It points at a proposal already on the table. let's may sit before the verb (ok lets go by the recommendations). A bare continue + task is a task and still scores;
    • opens with yes and is eight words or fewer, whatever follows (yes, dream then the hook). Nobody opens a new task with yes. ok and sure still need the verb: okay so the graph is broken reports a problem and scores;
    • is a status check of eight words or fewer opening with is/are/was/were/did/does/do/has/have
      • it/this/that/we/they/everything/all (is it done, did it work);
    • is a report of four words or fewer that work landed: merged, landed, closed, approved, deployed or done, alone or after all, both or PR numbers (16 merged, #541 is merged, 542 its merged);
    • asks only what is next, left or remaining (what is next, ok so what is left). It asks you to choose, so there is nothing to sharpen;
    • contains just, quickly, only, typo, or rename;
    • names an exact file path or file:line.

    Set CORTEX_NO_OPTIMIZE=1 to disable the optimizer entirely.

  2. Ask at most 2 questions, highest-value first, skipping any the prompt already answers: WHAT should happen (missing outcome), WHERE it lives (missing component), HOW it lands (new / change / migration). Ground every question in this repo's real names — "skills/ or tools/?" beats "which layer?".
  3. Synthesize one prompt: [ACTION] [COMPONENT] [in DOMAIN] [with CONSTRAINTS] -> [RITUAL]
  4. Confirm. Show it, wait for a one-word yes or an adjustment. One adjustment round, then go. Never act on an unconfirmed prompt.
  5. Save to docs/prompts/YYYY-MM-DD-<slug>.md — slug is kebab-case from 3–5 of the most specific words of the optimized prompt. If the file exists, append another block.
  6. Route to the ritual named in the synthesis.

Routing

Prompt shapeRitual
new feature or non-trivial changesuperpowers:brainstorming, then /design-spec if risky
bug, test failure, surprise behavior/debug-issue; superpowers:systematic-debugging where Cortex has no index of the repo
stray thought, link, task/quick-note, typed by the user — it is not model-invocable
vault structure or health/audit-notes
"make a ritual for X"/write-agent-docs

Record format

---
type: optimized-prompt
created: 2026-07-21
score: 4
ritual: superpowers:brainstorming
---
## Raw
<the original prompt, verbatim>

## Clarified
- Q: <question> -> A: <answer>

## Optimized
<the synthesized prompt>

Don't

  • Don't ask more than 2 questions. Friction is what gets this ritual disabled.
  • Don't fire on slash commands, confirmations, or mid-flow replies.
  • Don't act on a prompt the user hasn't confirmed.
  • Don't optimize a prompt that is already precise — silence is the correct output.
  • Don't write the record anywhere but docs/prompts/ (gitignored; keeps the privacy firewall).

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.