Owns: asynchronous decision questionnaire drafting.
Use this skill to turn a knowledge gap into a local Markdown draft for one explicit decision owner. Grill the send, not the subject: clarify who the user intends to ask, where the draft is intended to go, and what must come back. Do not run a live subject interview or try to answer the recipient's questions on their behalf.
This skill does not replace grilling. Use grilling when the
current user owns the decision and wants a live decision interview.
Preconditions
Require both:
- an explicit intended recipient: the named person or role that owns the missing knowledge;
- an explicit intended destination: the user-selected document path or later delivery channel where they plan to use the draft.
Never infer or select a recipient, destination, address, handle, or channel. If either is absent, do not write the questionnaire; ask one bounded clarification about the missing send detail and wait. Do not expand that clarification into a subject interview.
Also require a concrete decision or fact the user needs back and observable response criteria describing what a useful answer must cover.
Process
- Confirm the decision owner, intended recipient, intended destination, decision needed, and response criteria. Preserve the user's wording for names and destinations.
- Identify the gap between what the recipient knows and what the user needs. Draft only questions that close that gap; do not manufacture answers.
- Minimize and redact context before rendering. Include only the minimum context the
recipient needs. Replace credentials, tokens, cookies, personal data, and other
sensitive values with
<REDACTED>. Do not include conversation transcripts, unrelated command output, or private reasoning. - Order questions most important first. Use one decision or fact per question, an answer stub below each question, and a short reason only when the question could be misread.
- Render the draft to
to-questionnaire-<slug>.mdat the user-approved local path, using the exact structure below. Mark it as unsent. - Verify that every response criterion maps to at least one question, all identity and destination fields match the user's input, and no sensitive value remains.
- Return only the draft path, the intended destination label, and the unsent status. Do not paste the whole questionnaire back unless the user asks to review it.
Document structure
# <Questionnaire title>
Status: Draft — not sent
**Decision owner:** <person or role accountable for the resulting decision>
**Intended recipient:** <explicit person or role supplied by the user>
**Intended destination:** <explicit local path or later delivery channel supplied by the user>
## Purpose and decision needed
<what the user must be able to decide or do after the response>
## Context
<the minimum redacted context needed to answer well>
## Response criteria
- <observable fact, choice, constraint, or approval the response must provide>
## Questions
### <one decision or fact per question>
_Why this matters: <include only when clarification is useful>_
> <answer stub>
## Anything else?
What should we know that this questionnaire did not ask?
No-send boundary
Never send, post, email, upload, or publish the questionnaire. Do not call a connector, provider, messaging tool, or mail client. Do not open a compose window, choose a contact, or place the draft on a clipboard. Sending requires a separate explicit user action outside this skill, with the user controlling the final recipient, destination, and content.
Examples
- Valid: recipient
Example Recipient, destinationdecision-drafts@example.invalid, and decision "which migration window is acceptable." The file is rendered, but the draft remains local. - Missing destination: ask one bounded destination question and produce no file.
- The current user owns the answer and wants pressure-testing: use
grilling, not this skill.