grill-with-docs — relentless interviewing, grounded in reference docs
SkillMediaInterview the user relentlessly about a plan or design until reaching shared understanding, one question at a time, walking each branch of the decision tree, every question carrying a recommended answer, and every answer grounded in the project's own docs and examples rather than guesses. Use when the user asks to be grilled, wants a plan stress-tested, or when a design has to be pinned down before code is written. Built into a2ui-ask so agents that need the discipline have it even when no separate grilling skill is installed.
Available today. Use it from your connected AI after setup.
No other account needed.
Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.
Then ask your AI: use the grill-with-docs skill
What this skill tells your AI
The instructions your AI receives, as published by yuniqueunic/a2ui-ask in skills/grill-with-docs/SKILL.md and read by ahel’s review.
Interview the user relentlessly about every aspect of this plan until you reach a shared understanding. Walk down each branch of the decision tree, resolving dependencies between decisions one by one: a choice that constrains later questions gets asked before the questions it constrains. For every question, provide your recommended answer — the user should be confirming or correcting, not starting from a blank page.
Ask the questions one at a time, and stop when the answers have converged: no new branches are opening, and you could write the plan down without inventing anything.
If a question can be answered by exploring the codebase, git history, existing answers, or the reference docs below — explore instead of asking. The interview is for decisions only the user can make.
Ground every question in the docs
Before the interview starts, read the reference material the task points at — a feature brief should be shaped by how similar features are scoped in this repo, a UI by the design system, a schema by the engine's own examples. Concretely, for forms built with schemaui:
- the control cheat sheet and rich-content guide in the a2ui-ask
SKILL.mdthat ships next to this file; - the runnable forms in a2ui-ask's
examples/— the shortest path to "what does a good schema look like"; - schemaui's own controls gallery and rich-content gallery — every hint value and every figure surface side by side.
Reading beats guessing: an option list drafted after looking at the gallery is shorter and better labelled than one drafted from memory.
Interview discipline
- One question per turn. The user answers in one line; you follow up in the next. A wall of ten questions gets ten lazy answers.
- Recommend, then ask. "I'd go with X because Y — ok?" beats "what should we do about Z?".
- Resolve dependencies in order. Ask the question that narrows the space before the questions it narrows.
- Explore instead of asking whenever the answer is discoverable.
- Record what you heard. Restate the decision in one line before moving to the next branch, so a misunderstanding costs one message, not a whole plan.
- Stop when converged. Ending the interview is part of the discipline.
When the interview becomes a form
If the interviewer's questions are structured choices and config, stop asking in
chat: generate a draft-07 JSON Schema with your recommended answers as
defaults and hand it to a2ui-ask's ask script, so the user gets a real form
(see the a2ui-ask SKILL.md for the control cheat sheet). Use the form for the
cluster of decisions; keep the chat interview for the open-ended ones.
Signals
- GitHub stars
- 26
- Last commit
- Sep 2026
Advanced
- Item type
- skill
- Key
grill-with-docs-yuniqueunic- Source
- github.com/yuniqueunic/a2ui-ask