Clarify
My personal agent skills
npx -y skills add owainlewis/agent-skills --skill clarifyAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
What its author says it does
Copied from the file, not written here
Turns vague, messy, voice-dictated, or multi-part asks into a self-contained prompt a fresh agent can execute. Use when the user asks to clarify, stress-test, tighten a plan, or resolve references like "the thing". Interviews one question at a time, then outputs a final prompt.
SKILL.md
2.7 KB, as published. Nobody here has run it
Clarify
Turn a vague ask into a reusable Final prompt: block.
Always emit Final prompt: before doing work unless the user explicitly says just do it, just run it, or skip the prompt.
Workflow
- Read the input and discoverable context:
AGENTS.md,CLAUDE.md,README.md, file tree, relevant code. - Emit
Cleaned ask:only when the raw ask is noisy enough to obscure intent. - Ask one unresolved decision at a time.
- Include a recommended answer and one-line reason with each question.
- After each answer, re-check for new ambiguity.
- Stop asking when a fresh agent could execute the prompt without more context, or when the user says to write it.
- Emit one self-contained block under
Final prompt:.
Question format:
**Q:** <question>
**My recommendation:** <answer> - <one-line reason>
Ask About
- Inputs: file, glob, directory, required args.
- Outputs: location, naming, overwrite behavior.
- Failure modes: network errors, missing files, bad input.
- Scope edges: formats, environments, out-of-scope work.
- Hidden decisions: defaults the agent would otherwise invent.
- Success criteria: how the user knows it worked.
Do not ask about:
- Cosmetic choices.
- Project conventions the repo answers.
- Details already clear in the ask.
Final Prompt Contract
Include:
- Goal in one sentence.
- Exact file path or directory.
- Inputs and outputs.
- Dependencies, tools, env vars, and conventions.
- Failure behavior.
- Success criteria.
- Out of scope, when it prevents drift.
The prompt must read cold. No references to "what we discussed" or "the choices above".
Example Shape
Final prompt:
Build a Python script at `tools/auphonic/enhance.py` that takes one MP3 file
path as a required CLI argument and runs it through the Auphonic Simple API.
Auth: read `AUPHONIC_API_KEY` from the workspace `.env`.
Output: save `<basename>-enhanced.mp3` next to the input.
Overwrite existing output.
Failure: exit non-zero with a clear message for missing input, missing API key,
or API error.
Conventions: follow `tools/youtube/youtube.py` and use a PEP 723 uv header.
Out of scope: batch mode, presets, GUI.
Verify: run end-to-end against a real MP3.
Just Do It
When the user says just do it, state assumptions once, bake them into the final prompt, then act.