Context architectureStop re-explaining the same project rules to every AI tool.
Turn scattered requirements, architecture decisions and “never-do” rules into a small, versioned context layer that can travel across tools.
Time 30–90 min setupSkill IntermediateTools 1–3Human verification Required
The real problem
When work moves between research assistants, chat models and coding agents, project context fragments. Important constraints live in old chats, meeting notes and tribal knowledge. The result is repeated explanation, inconsistent output and accidental violations of project rules.
A portable context structure
# PROJECT CONTEXT
## 1. Mission
What this system exists to do:
[ONE PARAGRAPH]
## 2. Architecture
- Runtime:
- Languages/frameworks:
- Data stores:
- External services:
- Deployment:
- Important boundaries:
## 3. Non-negotiable rules
- Never:
- Always:
- Requires human approval:
- Security/privacy constraints:
- Performance constraints:
## 4. Domain rules
- Key entities:
- State transitions:
- Important invariants:
- Timezone/currency/locale rules:
## 5. Engineering conventions
- Testing:
- Error handling:
- Logging:
- Dependency policy:
- Naming / file conventions:
## 6. Current work
- Goal:
- In scope:
- Out of scope:
- Relevant files/services:
## 7. Decisions
For each important decision:
- Decision:
- Why:
- Alternatives rejected:
- Date / owner:
- Revisit trigger:
## 8. Unknowns
- Open questions:
- Assumptions that must not be treated as facts:
Workflow: messy inputs → stable context
01CollectGather PRD, architecture notes, README, meeting transcript, coding conventions and known operational constraints.
02ExtractSeparate durable rules from temporary task details. Durable rules belong in shared context; task details belong in the current request.
03Resolve conflictsDo not let AI merge contradictory requirements. Flag them for a human decision.
04VersionKeep the context file in source control where appropriate so changes are reviewable.
05Inject selectivelyGive each AI tool only the relevant context instead of dumping the entire organization into every session.
06RefreshRemove stale rules after architecture or process changes rather than letting context accumulate forever.
Prompt: convert raw documents into context
Convert the supplied project material into a durable AI context file.
RULES
- Preserve explicit architecture and business constraints.
- Separate durable project rules from temporary task details.
- Do not infer a rule just because it appears once in a discussion.
- If two sources conflict, show the conflict instead of choosing one.
- Label assumptions and unresolved questions.
- Do not invent framework versions, services, ownership or security requirements.
OUTPUT
1. Mission
2. Architecture
3. Non-negotiable rules
4. Domain rules
5. Engineering conventions
6. Current work
7. Decisions
8. Unknowns
SOURCE MATERIAL
[PASTE PRD / README / MEETING NOTES / ARCHITECTURE NOTES]
Important boundaryPersistent context should reduce repetition, not become a secrets dump. Keep credentials, tokens, customer data and unnecessary sensitive information out of broadly shared context files.