Grilling: Resolve a Decision Tree With the User

What it does

The source models a plan as a design tree. Each round asks the current frontier: questions whose prerequisite decisions are already settled. Recommendations make a short yes answer meaningful, but the human still owns the decisions. Environmental facts should be investigated rather than turned into unnecessary questions. Once answers arrive, the frontier is recomputed so dependent choices wait until their foundations are known. This is the reusable interview primitive used by other workflows; it differs from a generated one-shot questionnaire sent to another person.

When to use it

• Clarify scope before implementing an ambiguous feature.
• Stress-test operational decisions that depend on product choices.
• Expose assumptions hiding behind a seemingly simple project brief.

Step-by-step workflow

1. Identify the goal and map the decisions it depends on.
2. Separate discoverable environment facts from owner preferences.
3. Ask the currently unblocked questions with clear recommended answers.
4. Wait for real answers and update the decision tree for the next round.
5. Confirm the shared understanding before moving into execution.

Example request

Use grilling to clarify our customer export feature. Read the repository to find existing formats and limits instead of asking me to repeat them. Start with decisions that do not depend on unresolved answers. Recommend a default for each question, but do not invent my replies. Cover access permissions, failure behavior and scope. Stop for confirmation of shared understanding before implementation.

Worked example

The export format may determine whether streaming is required, while customer authorization rules can be discussed immediately. Asking a detailed CSV quoting question before choosing CSV creates unnecessary work. Conversely, existing code can establish the current maximum row count without an interview question. A good record distinguishes what the owner chose from what the agent found in the environment. When a preference changes, dependent decisions need reconsideration instead of remaining silently inconsistent.

What you get

A shared, explicit decision tree and a set of settled requirements suitable for the next planning step.

How to evaluate the result

• Verify later-round questions depend on settled prerequisites.
• Check facts are supported by inspection rather than assumed.
• Confirm the owner actually agreed the final understanding.

What you need

A present decision owner, access to relevant project facts and compatible exploration tools. The upstream primitive dispatches factual investigations to subagents; this is a client capability requirement, not permission granted by the catalogue.

Setup and activation

The publisher documents plugin setup for Claude Code, Codex and Copilot, and manual Agent Skills setup for Gemini CLI and other clients. Choose one distribution method per client to avoid duplicate skills. The engineering collection expects project-specific tracker, triage-label and document-location settings from setup-matt-pocock-skills. Keep referenced companion skills and files available. This catalogue review does not install or run them; scan the exact source before any later installation. Client tools, permissions and subagent support still need separate validation.

Limits and considerations

An exhaustive interview can be excessive for a small settled change. The agent cannot answer human-only choices on the user’s behalf. Research and implementation permissions remain separate from agreement on requirements.

Related workflows

• Use to-spec to capture an agreed engineering discussion.
• Use to-questionnaire when a different person holds the missing knowledge.

License and reuse

MIT at repository level (Copyright 2026 Matt Pocock). Preserve the copyright and permission notice when redistributing substantial source material; inspect any separately licensed dependencies or assets.

Further reading

• Reviewed SKILL.md (immutable commit)
• Publisher README: distributions and compatibility
• Reviewed MIT license