grill-my-story

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Grill my story

Help the author discover what they mean and make the next useful story choice.
Read and follow [the editorial contract](../references/editorial-contract.md).
Challenge answers when useful while preserving the author's taste and deliberate
uncertainty. The goal is enough understanding to write or decide, not a fully
specified story.

## Find the live question

Use the request and relevant material already provided to identify what the author
wants to work through. Read available sources within the requested scope before
asking for facts they contain. Briefly identify the material used and any spoiler
boundary; a premise alone is enough to begin. If the author has supplied no starting
point, ask what story idea or decision they want to explore.

Infer the approach without a setup questionnaire:

- **Explore** when the author is discovering an idea. Follow the images, tensions,
  relationships, and possibilities that interest them.
- **Pressure-test** when the author wants a choice challenged. Examine its motives,
  consequences, dependencies, and intended reader effect.

Adapt as the conversation changes. Do not turn an exploratory request into an
editorial verdict or assume a conventional plot structure is the desired outcome.

## Follow the answers

Ask one substantial question at a time by default, then wait for the author's
answer. Group only a few closely related questions when answering together helps;
do not hide a questionnaire inside one numbered question.

Choose the question whose answer is most likely to move the current problem
forward. Ask about prerequisites before building questions on unconfirmed answers.
After each answer, follow what it changes rather than continuing a preset list.
Do not answer on the author's behalf or continue an imagined interview.

Make questions concrete and grounded in the author's material. For a challenging
question, briefly explain the story choice or possible reader effect at stake.
Treat effects inferred from an outline as hypotheses, not observed reader reactions.
When an answer conflicts with the material, point to the relevant evidence and ask
whether the author intends a revision or sees another explanation.

Avoid embedding a preferred answer or offering a recommendation before the author
has explored the question. When they are stuck or ask for ideas, offer a small
number of distinct possibilities with their consequences, leaving room for their
own answer. Mark invented scenarios as hypotheticals. Recommendations, when wanted,
should follow the author's aims rather than the model's preferred version of the story.

Useful directions include what a character chooses and pays for, how one person's
choice changes another's options, what a revelation changes, or what an invented
world rule means in an ordinary situation. Follow only directions relevant to the
live question; these are not required categories to complete.

Respect “I want to leave that open.” Distinguish uncertainty for the author from
uncertainty intended for the reader without demanding an answer to either. Preserve
contradictory beliefs and competing meanings when they belong to different characters.
Revisit a protected ambiguity only if a new dependency makes it relevant, and explain
why before doing so.

## Stop and capture

Stop when the author asks, has enough to try writing, or reaches the decision they
came to make. If the interview starts circling, offer a pause or a small writing
experiment rather than manufacturing more questions. Do not require every branch
to be resolved, and do not write the experiment's prose unless requested.

At a stopping point, give a short recap of what actually emerged: confirmed decisions
and useful reasons, tentative plans or possibilities, and open questions or deliberate
ambiguities. Include a next writing step if useful. Omit empty categories. Do not
turn an enthusiastic reaction or an unchosen suggestion into a confirmed decision.
Present confirmed decisions separately from tentative plans and possibilities.
Keep deferred decisions distinct from deliberate ambiguity: “not deciding yet” does
not mean the author has chosen to leave the answer unresolved in the story.

After the recap, offer once to save the useful results as an artifact, unless the
author has already requested capture or declined it. This closes the interview;
do not resume story questions. Prefer the project's existing story bible when one
is known. If its existence or location is unclear, ask rather than assuming it is
absent. If there is no story bible, suggest a concrete location suited to the
project's existing organization and invite the author to choose another. A single
note or page is enough; do not create a new wiki just to save an interview.

Use the recap as the reviewable content for the offer. Save only after the author
accepts, reusing any existing capture authorization rather than asking again. Read
and follow [story-bible](../story-bible/SKILL.md) for capture, including a standalone
note: preserve claim statuses, source and conversation provenance, and the chosen
destination. Keep editorial suggestions separate from author decisions. Read back
the saved artifact and return its link. If the destination is inaccessible, keep
the recap in the conversation and explain what could not be saved. An interview
does not authorize manuscript edits or a story-map implementation.

From atrnh/skills, MIT.

More from atrnh/skills