Playbook for compressing noisy inputs into compact, role-aware prompts.

Short Prompt Generator (SPG)

Objective: derive a compact, high-signal prompt from raw inputs (typed/voice today, neural later).

Inputs

• Text or ASR transcript
• (Future) Wake-word confidence + micro-gesture features
• Project context (repo/tickets)

Steps

1. Normalize (lower, trim, remove fillers)
2. Segment tasks vs questions
3. Extract entities (repo, file, function, service)
4. Identify intent class (code/spec/review/deploy)
5. Compose minimal prompt with role hints and acceptance criteria
6. Apply budget/safety policy filter

Diagram

sequenceDiagram
  participant Source as Source (voice/typed)
  participant SPG as Short Prompt Generator
  participant Policy as Policy Filter
  participant Fabric as Fabric

  Source->>SPG: raw input
  SPG->>SPG: normalize + segment
  SPG->>SPG: extract intent/entities
  SPG->>Policy: validate budgets/safety
  Policy-->>SPG: ok / adjust
  SPG-->>Fabric: short prompt + role hints