Pattern Proposal: Steering Projection and Prompt Fact Synthesis

Summary

Compiles raw architectural deltas and capability lineages into compact, typed PromptFact objects for LLM context steering.

Candidate Pattern

• ID: steering-projection-fact-synthesis
• Classification: intelligence
• Maturity: established
• Type: implemented

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/steering/models.py (PromptFact, SteeringRuntimeIndex, CapabilityLineage, ArchitecturalDelta) - Defines typed entities that distill git changes into structured steerable context.

Scope Assessment

• Evidence Strength: high
• Confidence: high
• Applies to: ideas-to-life-cas

Rationale

• Intent: Prevent LLM architectural drift without overloading context windows with redundant history.
• Context: Used during pull request evaluation and workflow runs in CAS runner before invoking agentic synthesis.
• Force: Repository history is too large to fit in raw LLM context
• Force: Agents make poor design decisions when unguided by domain rules
• Force: Rules must update dynamically as code matures

Traceability

Artifact source: proposed-patterns