writing-interview
Skill0 starsChanged 25 days agoMITView on GitHub
Interview Shane before drafting a piece, so the draft is built from his own material instead of generic model prose. Use when Shane says: "interview me first", "help me write about X", "let's write a post on X", "I want to write about", "/writing-interview", or when a new piece is being started with no source material supplied. Runs before drafting and writes no prose until the interview is answered.
In this repository5 other files it gives its agents.
- humanizeSkill
- ms-style-passSkill
- plain-languageSkill
- ste100Skill
- writing-reviewSkill
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
- Distinctive: line 19
Do not draft before Shane answers: Not an opening paragraph, not an outline presented as settled, not "here's a rough angle to react to." …
- Distinctive: line 21
If the request genuinely needs no personal material — a doc rewrite, a definition, a status note — this is the wrong skill.
- Distinctive: line 49
Do not interview about the whole thing.
- Distinctive: line 62
Add your own prose only for a clear job: sourced research, a definition, a factual bridge, or compression.
- Very specific: line 63
Leave [TK: precise question] where the material is still missing.
- Distinctive: line 74
Put this in chat, not in the draft file.
≈ “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.
Files
Writing Interview
One job: collect the material only Shane has, before anything gets drafted.
Why this exists
Profiles/voice.md's do_not_infer closes on the rule that makes this skill necessary: "Don't generate Shane-voiced content without grounding in my actual thoughts or notes. The voice is portable. The thinking isn't."
A model can reproduce his cadence from the profile. It cannot reproduce what he saw, measured, decided, or changed his mind about. When a piece needs that material and nobody supplied it, the draft invents something that reads like it. So the interview happens first, and drafting waits.
The hard gate
Do not draft before Shane answers. Not an opening paragraph, not an outline presented as settled, not "here's a rough angle to react to." Ask, then wait.
If the request genuinely needs no personal material — a doc rewrite, a definition, a status note — this is the wrong skill. Say so and hand off rather than running a ceremony.
What to ask
Open with one prompt, not a questionnaire. Ask for a single untidied take:
Talk to me for three to five minutes. Don't organise it. What happened that made you want to write this, who is it for, what do you believe about it, and what examples or doubts make the idea yours?
Then use these as follow-ups, not a list to work through:
- What triggered this — a conversation, a bug, a result, an annoyance? The trigger, not the topic.
- Picture one reader. What do they already know, and what should change for them after reading?
- Who disagrees, and what is their strongest real argument?
- What do you say about this out loud that you have never written down?
- Which example from your own work carries the point? What happened, and what changed?
- Where are you uncertain, or what have you changed your mind about?
- What would you cut from the conventional version of this piece?
Also sweep the vault for prior material: obsidian search on the topic, then obsidian read the best hits. Existing notes and posts are source material and carry his actual language.
The question that works hardest
"What would you cut from this that everyone else would keep?" tends to produce more usable material than any other prompt.
One concrete follow-up beats five general ones: "Tell me about the last time this actually bit you. What broke, and what did you do?"
If a draft already exists
Do not interview about the whole thing. Ask at most three questions, aimed only at the paragraphs that lack support or authorship:
This paragraph says [claim]. What is your evidence, and how sure are you?
The example could belong to anyone. What is your version of it?
The ending restates the point. What should the reader actually do, notice, or reconsider?Turning answers into a draft
- Extract the point, the useful phrases, the examples, the uncertainties, and the order he discovered them in.
- Build the spine from his language. Cut repetition, and cut anything addressed to you rather than to the reader.
- Edit grammar lightly where comprehension needs it. Protect coined language, trade slang, genuine asides, mixed feelings, and unusual details. Where a change would alter the thought or erase a distinctive phrase, keep the original or show both.
- Add your own prose only for a clear job: sourced research, a definition, a factual bridge, or compression. Never write a memory, preference, or experience for him.
- Leave
[TK: precise question]where the material is still missing. - Report the provenance block, outside the draft.
Provenance
Author material: <which sections or language came from the interview or prior writing>
Model contribution: <research, organization, connective prose, or none>
Open items: <the [TK] list, or none>Put this in chat, not in the draft file. Do not report a detector score as evidence of authorship. The provenance note describes what actually happened, and that is the claim worth defending.
Handoff
Output is a draft ready for workflow-stage-draft --resume-from humanize, with the provenance block and the open [TK] list already assembled. publish-post refuses to ship while any [TK] is unresolved.
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
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.

