grill-with-docs

name	grill-with-docs
description	Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.

Effort: This is a deep design-questioning session — run it at xhigh effort (output_config: { effort: "xhigh" }) per the effort-level mapping in CLAUDE.md. budget_tokens is deprecated on Opus 4.8+; use thinking: { type: "adaptive" } with an effort level instead.

Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.

Ask the questions one at a time, waiting for feedback on each question before continuing.

If a question can be answered by exploring the codebase, explore it via a subagent, not in this thread — see "Staying in the smart zone" below. The main grilling thread must stay lean; it sees distilled conclusions, never raw file dumps.

Staying in the smart zone

This is a long, interactive, HITL session. Left unbounded it accumulates file dumps, every doc read up front, and the entire verbatim Q&A — pushing the main thread out of the smart zone (sharp reasoning) into the dumb zone (a filling context window degrades reasoning) right when the hardest design questions arrive. Butler's CLAUDE.md makes this a rule:

Smart Zone: Keep session context under 100k tokens. When approaching the limit, use /handoff to create a continuation document and start a fresh session.

Phase 1 grilling is exactly where this matters most. Apply these levers throughout the session:

1. Offload exploration to a subagent (primary lever)

When a branch needs codebase or doc exploration to answer, dispatch a subagent (e.g. the Explore agent, or a general-purpose Agent) and have it return a distilled summary — the conclusion, the relevant term, the one contradicting line — not the files. Raw file contents stay in the subagent's context and never enter the grilling thread. Spawn several in parallel when branches are independent.

Ask the subagent for exactly what the current question needs ("Does the bookings code cancel whole Orders or partial? Return the function and a one-line answer"), not "summarise the codebase".

2. Read docs scoped, on demand

Do not read all of CONTEXT.md and every ADR up front. Pull only the section relevant to the branch in front of you, when you reach it. Let the offload subagent grep for the specific term/decision rather than loading whole documents into the main thread.

3. Compact resolved branches

Once a decision is captured (in CONTEXT.md or an ADR), the verbatim back-and-forth that produced it is no longer needed in-thread. Summarise a resolved branch to its outcome ("✓ 'cancellation' = partial, per ADR 0007") and move on; don't keep re-citing the full exchange. The durable record lives in the docs, not the conversation.

4. Smart-zone checkpoint

Watch the context footprint. As it approaches the ~100k threshold from CLAUDE.md, stop and prompt the user to /handoff — write a continuation document capturing the decision tree resolved so far and the open branches, then resume the grilling in a fresh session. Never silently grind past the threshold into the dumb zone; surfacing the checkpoint is part of the skill, not an interruption to it.

Domain awareness

During codebase exploration (offloaded per lever 1 above), also look for existing documentation:

File structure

Most repos have a single context:

/
├── CONTEXT.md
├── docs/
│   └── adr/
│       ├── 0001-event-sourced-orders.md
│       └── 0002-postgres-for-write-model.md
└── src/

If a CONTEXT-MAP.md exists at the root, the repo has multiple contexts. The map points to where each one lives:

/
├── CONTEXT-MAP.md
├── docs/
│   └── adr/                          ← system-wide decisions
├── src/
│   ├── ordering/
│   │   ├── CONTEXT.md
│   │   └── docs/adr/                 ← context-specific decisions
│   └── billing/
│       ├── CONTEXT.md
│       └── docs/adr/

Create files lazily — only when you have something to write. If no CONTEXT.md exists, create one when the first term is resolved. If no docs/adr/ exists, create it when the first ADR is needed.

During the session

Challenge against the glossary

When the user uses a term that conflicts with the existing language in CONTEXT.md, call it out immediately. "Your glossary defines 'cancellation' as X, but you seem to mean Y — which is it?"

Sharpen fuzzy language

When the user uses vague or overloaded terms, propose a precise canonical term. "You're saying 'account' — do you mean the Customer or the User? Those are different things."

Discuss concrete scenarios

When domain relationships are being discussed, stress-test them with specific scenarios. Invent scenarios that probe edge cases and force the user to be precise about the boundaries between concepts.

Cross-reference with code

When the user states how something works, check whether the code agrees (via an offload subagent — lever 1). If you find a contradiction, surface it: "Your code cancels entire Orders, but you just said partial cancellation is possible — which is right?"

Update CONTEXT.md inline

When a term is resolved, update CONTEXT.md right there. Don't batch these up — capture them as they happen. Use the format in CONTEXT-FORMAT.md.

CONTEXT.md should be totally devoid of implementation details. Do not treat CONTEXT.md as a spec, a scratch pad, or a repository for implementation decisions. It is a glossary and nothing else.

Offer ADRs sparingly

Only offer to create an ADR when all three are true:

1. Hard to reverse — the cost of changing your mind later is meaningful
2. Surprising without context — a future reader will wonder "why did they do it this way?"
3. The result of a real trade-off — there were genuine alternatives and you picked one for specific reasons

If any of the three is missing, skip the ADR. Use the format in ADR-FORMAT.md.