draft-distill-proposals

Extracting learning signals from a merged PR diff involves analyzing implementation changes alongside optional issue STM evidence to detect targeted updates. This process generates staged proposals for future review.

Proposal staging in enterprise agentic workflows is the process of writing detected learning signals to an issue STM evidence area. This keeps product LTM read-only while allowing optional human review before final capture.

Generating a proposals.yaml schema from PR evidence requires analyzing merged diffs and STM artifacts to classify learning signals. The output is a forward-compatible file capped at two proposals to avoid trivial noise.

Deterministic development does not strictly require STM evidence for learning extraction, but reading available issue evidence raises or lowers confidence. This helps prioritize the strongest learnings from the PR diff.

Limitations of automated PR diff learning extraction include capping output at exactly two staged proposals and avoiding trivial-noise outputs. The system maintains a read-only product LTM to prevent unauthorized updates.

You can use issue STM artifacts to raise or lower confidence in staged proposals by providing enhance and fix-it evidence. This STM-evidence-aware approach prioritizes the strongest learnings from the diff.