infer-epics-from-code

Epic inference from code works by clustering inferred capabilities using git co-change signals, knowledge-base domain groupings, version tags, and ADR titles. This mechanism grounds the epic clustering process in actual repository evidence.

You generate epic proposal stubs by running an inference process that produces one YAML stub per derived epic. It creates ICE-schema artifacts and a decision manifest while recording resolution-trace probes for downstream enrichment.

Yes, epic inference requires existing scan-index evidence and upstream proposals. Specifically, it runs after infer-features-from-code has produced features.yaml and enriched-capabilities.yaml to ensure boundaries are evidence-driven.

When requirements cannot be inferred from code evidence, the system emits knowledge-gap placeholders. It avoids fabricating expectations by marking failure scenarios and business rules as knowledge gaps for stakeholder review.

The best way to bootstrap milestones for a repository with missing structure is to infer minimal, human-readable epic boundaries from code co-change and ADR titles. This allows teams to start planning without getting blocked by missing documentation.