Builds a new readvisor out of a real person's judgment, by interview, and installs it. Two ways to gather: LIVE — a patient interview of the user themselves, one or two questions at a time, twelve to eighteen in all; or QUESTIONNAIRE — a plain, email-ready file the user sends to someone whose advice they want on hand (a friend, a mentor, a colleague, a parent, a former editor), with the answers pasted back whenever they arrive. Always ends the same way: a follow-up pass to the user, a preview of both documents they can correct line by line, and their go-ahead before enough installs anything. Produces AGENT.md and MOTIVATION.md and installs them with the install_readvisor tool, in this project or on this machine. Use for 'make me a readvisor', 'build a readvisor', 'new readvisor', 'readvisory', 'interview me and turn it into a readvisor', 'I want my friend's judgment in here', 'profile how my old editor thinks', 'send my dad a questionnaire', 'turn these answers into a readvisor', 'I want a second opinion that argues like X', 'someone who will push back on me about Y', 'I miss asking her about this'. Not for renaming the chief readvisor, not for running a council — this makes the participants, it does not convene them.
A readvisor is a second opinion with a particular judgment behind it. This skill makes one, from a real person, by asking them questions and writing down what their answers reveal about how they think.
The person can be the user, interviewed live. Or it can be someone else, who
answers a questionnaire in their own time and sends it back. Either way the
work ends in the same two documents — AGENT.md and MOTIVATION.md — and
enough installs them as a readvisor the user can switch on.
This is a favor, not a form. Somebody is spending their attention so that somebody else can keep asking them things. Be a trusted collaborator, not an interrogator. Warm, curious, unhurried. If they joke, laugh. If they go quiet, slow down. If they correct you, take it.
rness/policies/requests.md. This job is
multi-turn by construction and often multi-day — a questionnaire can sit
in somebody's inbox for a week. The request file is the tracker that lets
a later session pick the work up cold. Show the user its path.references/privacy.md. One page. It sets what you keep, what
you abstract, and what you never ask about. Read it before the first
question, not after the first slip.| Step | What happens | Load |
|---|---|---|
| 1. Brief | What will this readvisor readvise on, whose judgment, what is it called | references/question-bank.md § brief-shaping |
| 2. Questions out | 12–18 questions, live or as a questionnaire | references/question-bank.md, references/coverage.md, references/questionnaire.md |
| 3. Answers in | Saved verbatim | — |
| 4. Follow-ups | 3–6 questions to the user, always | references/followups.md |
| 5. Synthesis preview | Both documents drafted, shown, corrected | references/synthesis.md |
| 6. Install | install_readvisor, project or global | this file |
Do not skip step 4. It is the step that makes an enough readvisor better than a transcript.
Three things, in plain conversation. Two or three turns, not a form.
block-breaker, careful-reader, aunt-rosa;The Block-Breaker, Careful Reader, Aunt Rosa.
Suggest both from what they have told you; let them overrule.references/question-bank.md opens with four brief-shaping questions that
turn a vague subject into a sharp one. Use them when the brief is soft.
Write the brief into the request file before going further.
Choose 12–18 questions from references/question-bank.md, using the
coverage checklist in references/coverage.md to decide which. Shape them to
the brief: a readvisor about cutting drafts needs different questions than
one about hard conversations.
One or two questions at a time. Never a flood. Let each answer change the
next question — that is the whole point of doing it live.
references/coverage.md has the next-question checklist; it is short, and
you should follow it literally.
Between questions, be a person: react to what they said, say what you are hearing, ask "why" when the why is where the signal is.
Write the whole set as one self-contained file:
rness/io/output/readvisory/<name>/questionnaire.mdreferences/questionnaire.md gives the exact shape — greeting, why they are
being asked, how long it takes, numbered questions with lettered choices, and
how to reply — plus a complete worked example. Follow it closely; the file
leaves the machine and you will not be there to explain it.
Then:
**Status:** waiting-on-user and fill the "Open
questions" line under Continuation: what is awaited, from whom, and where
the answers should land when they arrive.The answers arrive one of two ways:
rness/io/input/. read_file it — any format
enough converts is fine, the twin comes back as markdown.Save them verbatim, exactly as written, to:
rness/io/output/readvisory/<name>/answers.mdVerbatim matters. The abstraction happens in the synthesis, not here — you
will want the person's own phrasing later, and a paraphrase made now cannot
be un-made. references/privacy.md says what this does and does not mean.
Answers come back messy: out of order, three questions skipped, two answered in one paragraph, a long story where a letter was expected. All fine. Read them as a whole before deciding anything is missing.
The respondent's answers say how that person thinks. They cannot say what the user needs from them. That is a separate interview, and it is short.
Ask the user 3–6 questions: how they know this person's judgment, where it has been right and where it has been wrong, what they want to be challenged on, where this readvisor should defer to others, and what tone they want from it.
When the user interviewed themselves in step 2, this pass becomes the contradiction-and-stress round instead — the same slot, different questions.
references/followups.md has both versions, with examples, and says which
part of which document each answer feeds. Save the answers to:
rness/io/output/readvisory/<name>/context.mdDraft both documents. Show them in chat before writing anything final.
references/synthesis.md is the reference for this step: the exact section
headings, the voice rules, how to mark an interpretive leap, how to write an
honest hand-off, what to do when the signal is thin, and the correction loop.
Read it fully — this is the step where a readvisor either sounds like a
person or like a personality test.
Invite three kinds of correction, by name:
Iterate until the user signs off. Keep the working drafts at
rness/io/output/readvisory/<name>/AGENT.draft.md and
MOTIVATION.draft.md so a reset does not lose them.
Ask the user one question: project or global?
rness/readvisors/<name>/, this project only, on in this
project. Right for a readvisor tied to one piece of work.~/enough/readvisors/<name>/, available in every project on
this machine, on here and off by default elsewhere. Right for a person's
judgment the user will want again.Then call the tool once, with the full text of both documents:
<tool name="install_readvisor">
<name>careful-reader</name>
<scope>global</scope>
<display>Careful Reader</display>
<agent_md>
# Careful Reader
You are a readvisor built from a working newspaper editor's judgment
about when a piece of writing is finished…
---
enough-tooltip-text: "careful-reader tells you whether a draft is done, and what it is still owed."
</agent_md>
<motivation_md>
# Motivational Substrate — Careful Reader
The values and frame beneath the advice…
</motivation_md>
</tool>Every field is required except display, which defaults to the folder name.
enough validates the name, checks both documents are non-empty and under
40 KB, requires the trailing enough-tooltip-text: line on agent_md, and
runs the same payload scan it runs on any untrusted skill. If it refuses, it
says exactly why — fix that and call again. Do not write the readvisor's
folder yourself with write_file; the tool is the only door.
Installing an existing name needs <replace>yes</replace> as an extra field,
and enough will still refuse to replace a readvisor that shipped with it.
Afterwards: fill in "End output" in the request file (both paths, the scope,
the display name), set **Status:** waiting-on-user, and tell the user it is
in the sidebar. Never move the request file to done/ yourself — that
move is the user's approval act.
rness/io/output/readvisory/<name>/
questionnaire.md the file the user sends out
answers.md what came back, verbatim
context.md the user's follow-up answers
AGENT.draft.md working draft
MOTIVATION.draft.md working draftInstalled readvisors land in rness/readvisors/<name>/ or
~/enough/readvisors/<name>/, written by the tool. The working folder stays
where it is — it is the record of how this readvisor came to exist, and the
starting point if the user ever wants to deepen it.
MOTIVATION.md says so. A thin readvisor,
honestly written, is more useful than a thick one that is making things up.references/privacy.md governs what
reaches the documents. When in doubt, abstract one level and say that you
did.enough-tooltip-text: "use readvisory to build a new readvisor out of someone's real judgment — interview yourself live, or send a plain questionnaire to the person whose advice you keep wishing you could ask — and enough installs the result in this project or on this machine."
0gsd/enough
3cf8a60
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.