grill-with-docs

Trigger phrases

Phrases that activate this skill when typed to Claude Code:

• grill with docs
• stress-test against the docs
• grill this against CONTEXT.md

What it does

grill-with-docs is the project-aware sibling of grill-me. Where grill-me interrogates a plan in isolation, grill-with-docs interrogates it against the documents your project already has — domain glossary, decision records, brand book, CLAUDE.md, whatever’s there. Each question is grounded in something the project already says.

The result: sharper plans plus a continuously-updated CONTEXT.md (or BRAND.md, or whichever doc you’re maintaining) that absorbs new terminology as the session resolves it.

It’s the difference between abstract grilling and grilling that respects what’s already been decided.

When to use it

• Working in a project that already has a glossary, ADR folder, or CONTEXT.md
• A new plan that uses overloaded terminology (e.g., “user,” “session,” “agent”) where consistency matters
• Onboarding a feature into an existing domain — you don’t want to redefine concepts that already have names
• Whenever drift in vocabulary between docs and code is becoming a real cost

When not to reach for it:

• Greenfield projects with no docs yet — use plain grill-me instead
• Throwaway prototypes where consistency doesn’t yet matter
• Plans where the docs themselves are wrong and need to be rebuilt — fix the docs first

Install

Distributed via Pocock’s skills repo. Install via his recommended path (npx skills add or manual SKILL.md copy into .claude/skills/grill-with-docs/).

The skill expects to find a CONTEXT.md (or similar) in the project root, or it will offer to create one as part of the session.

What a session looks like

A typical session has four phases:

1. You state the plan. Sloppy is fine.
2. The skill scans the project’s docs — CONTEXT.md, BRAND.md, decision records, root README — and indexes the terminology.
3. It picks one branch and asks one question, citing the doc. “Your plan uses ‘session’ but CONTEXT.md defines it as a 30-minute window. Did you mean that, or something else?” — narrow, grounded, sharp.
4. As you answer, the skill updates the doc inline to absorb new terms or refine existing definitions. By the end of a session, your plan is sharper and your CONTEXT.md is more accurate than it was an hour ago.

The discipline that makes it work: questions are grounded in actual document content, not generic best practices.

Receipts

Where it works well:

• Catches semantic drift between plan language and project glossary before it metastasizes
• Forces precision on overloaded terms (the words you’ve been using interchangeably for months)
• Produces a side-effect that compounds: every session leaves the docs slightly better

Where it backfires:

• If the docs are themselves vague or contradictory, the skill grinds — fix the docs first
• On a project with no docs, the skill offers to create one but the session feels less load-bearing than grill-me
• Over-grilling on small features can produce noise rather than signal

Pattern that works: invoke it before any feature work where terminology consistency matters across code, docs, and conversation. The session takes 10–20 minutes and the doc-update side-effect is often more valuable than the plan itself.

Source and attribution

Originally written by Matt Pocock. Canonical SKILL.md and supporting files live in the grill-with-docs folder of his public skills repository.

License: MIT. You can install, adapt, and redistribute with attribution preserved.

For the formal spec, defer to the source repo. This page documents the skill from a practitioner’s perspective.