| name | doc-grounded-grilling |
| description | Stress-test a plan, feature, or design against the repo's existing docs, domain language, ADRs, and context files. Use when a task is ambiguous, terminology is fuzzy, architecture trade-offs matter, or the user wants alignment before implementation. Ask one question at a time, inspect the codebase when answers are already present, and sharpen the project's language as decisions crystallize. |
| license | MIT |
| metadata | {"abvx_status":"experimental","abvx_origin":"adapted"} |
Doc Grounded Grilling
Use structured questioning to align a plan with the repo's actual language, constraints, and existing decisions.
Workflow
- Read the nearest relevant context first:
AGENTS.md
docs/ai/*
- ADRs
- repo docs
designmd-brand-kit artifacts when frontend or brand work is involved
- Ask one question at a time.
- If the answer is already in the code or docs, verify there before asking.
- For each question:
- identify the ambiguity;
- give a recommended answer or framing;
- wait for user feedback before continuing.
- Walk down the branches of the design tree until:
- terms are precise;
- boundaries are understood;
- key trade-offs are named;
- critical scenarios are covered.
- When a term or decision settles, classify the durable artifact:
- glossary/context update for canonical domain language;
- ADR candidate for hard-to-reverse or surprising trade-offs;
- PRD/spec note for scope and acceptance criteria;
- no artifact when the answer is temporary or already documented.
What To Challenge
- terminology that conflicts with existing domain language;
- assumptions contradicted by code or ADRs;
- vague nouns like "account", "session", "client", "job", "draft", "campaign";
- plans that skip the test seam, integration seam, or rollout seam;
- frontend ideas that ignore the current design system or brand shape.
Output
The session should leave behind:
- sharper language;
- clarified decisions;
- explicit open questions;
- identified docs that should be updated through
durable-context-maintenance when the decision settles;
- proposed glossary/context or ADR updates when the task changed the repo's shared language or a consequential decision.
Guardrails
- Do not interrogate for its own sake; stop when the plan is clear enough to execute.
- Do not invent domain language when the repo already has one.
- Do not batch ten questions at once.
Final Report
Summarize resolved terms, decisive trade-offs, remaining open questions, and which docs should be refreshed next.