← Previous: Being active, not passive · README · Next: Using smart models for grilling
The asset you are building
- While answering questions you are growing a context window full of answers you gave and design decisions you took — that context is incredibly valuable (
6:55). - The failure mode: people don't value what they created during the grilling session (
6:55).
Decision rule at the end of a grill
| Situation | What to do |
|---|---|
| Enough context budget left | Implement immediately in the same session, off the design decisions already made — no handoff needed (6:55, 7:21) |
| Budget exhausted / must quit out | Make a PRD — the two PRD skill produces an engineering-tailored handoff document, useful multi-session or even single-session (7:21) |
The "crazy one"
- Some people clear the context first, open a new context window, and only then run the PRD skill (
7:45). - That throws away a session holding on the order of 100,000 tokens of good design decisions — the author calls this totally crazy (
7:45).
flowchart TD
S["Grilling session<br/>~100k tokens of design decisions"] --> Q{Context budget left?}
Q -->|Yes| I[Implement in place]
Q -->|No| PRD["Run two PRD *in this session*<br/>→ handoff document"]
Q -.->|WRONG| X["Clear context, then run PRD<br/>❌ decisions are gone"]