Pattern Proposal: Steering Fact Projection

Summary

Compiles raw architectural repository metadata and lineage records into discrete, prompt-ready fact objects for dynamic agent context injection.

Candidate Pattern

• ID: steering-fact-projection
• Classification: intelligence
• Maturity: strategic_direction
• Type: emerging

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/steering/models.py (PromptFact, SteeringRuntimeIndex, CapabilityRecord, CapabilitySet) - Declares explicit fact projection models used for prompt rendering.
• .github/workflows/cas-build-steering-projection.yml - CI workflow executing steering projection compilation.

Scope Assessment

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

Rationale

• Intent: Provide agents with verified ground-truth context without needing ad-hoc embeddings or uncontrolled filesystem traversal.
• Context: Used during pipeline initialization to project governance and steering expectations into the runtime memory of the agent.
• Force: Passing raw markdown or git histories directly consumes excessive context tokens.
• Force: Unstructured context injection increases the likelihood of model hallucination regarding current architectural constraints.

Traceability

Artifact source: proposed-patterns