Actively build and sharpen the project's domain model as you design. This is the active discipline — challenging terms, inventing edge-case scenarios, and writing the glossary and decisions down the moment they crystallise. (Merely reading CONTEXT.md for vocabulary is not this skill — that's a one-line habit any skill can do. This skill is for when you're changing the model, not just consuming it.)
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.
Design awareness
If the work touches UI, the visual system has its own glossary: DESIGN.md. It is to design what CONTEXT.md is to the domain — the settled conventions (color, type, spacing, component inventory, layout patterns), not the reasoning behind them. Read it if present and challenge the plan against it the same way you do CONTEXT.md.
Create DESIGN.md lazily — only when the first visual convention is settled. If the repo already keeps a design-system doc, update that instead of inventing your own. Use the format in DESIGN-FORMAT.md. Keep it to settled conventions — the why behind a hard visual decision goes in an ADR, not here.
Roadmap awareness
When the work is one slice of a larger ambition, the plan above it lives in ROADMAP.md: the higher goal and the areas still ahead, each a future PRD. Read it if present and challenge the plan against it the way you do CONTEXT.md. As a broad scope narrows to the slice you're building, record the branches you're deferring there so they aren't lost — they seed the next PRDs.
When a grill reshapes the higher goal — a new area, a reprioritisation, a slice that's turned out wrong — revise ROADMAP.md inline, the same way you update a term whose meaning shifted. If the change moves the product or vision itself, push it up into the PRODUCT.md/VISION.md it defers to, not just the roadmap.
Create ROADMAP.md lazily — only when the scope is clearly bigger than one PRD. If the repo already keeps a product or vision doc (PRODUCT.md, VISION.md), update that instead of inventing your own. Use the format in ROADMAP-FORMAT.md.
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. 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.
Record resolved decisions inline
When a question resolves into an implementation-relevant answer — a constraint, negative requirement, edge case, numeric default, or ordering decision — append it to the feature's decision ledger right then, the same way you capture terms in CONTEXT.md. This is the durable record to-prd and to-issues check against, so the precise answer survives downstream instead of being softened into "persist sessions" or "support retry". Use the format in DECISIONS-FORMAT.md.
The ledger is not the glossary, a design convention, or an ADR — it captures what the feature must do, verbatim. It is a deliverable: commit it.
Offer ADRs sparingly
Only offer to create an ADR when all three are true:
- Hard to reverse — the cost of changing your mind later is meaningful
- Surprising without context — a future reader will wonder "why did they do it this way?"
- 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.
New ADRs created during planning should usually be proposed, not accepted — promote them only after the implementing work lands. If you're recording a decision the current codebase already reflects, mark it accepted.