RepoSense provides a layered, auditable path from source code to engineering insight:
- Analysis
- Learn
- AI Insights
- Studio
Code Repo
-> Analysis (Findings, Events, Evidence)
-> Facts artifacts
-> Pattern Engine (deterministic)
-> Insights (summary, risks, explain, ask)
-> Studio / Learn UI surfaces
- Analysis is the extraction layer. It produces Findings, Events, Evidence, and structured facts artifacts.
- Learn is the grounded learning layer. It organizes Concepts and Cases and links back to Evidence.
- AI Insights is the derived reasoning layer. It consumes Facts + Patterns first, then optionally uses constrained source drill-down.
- Studio is the local read UI layer for run outputs, risks, explain, snippets, and links into Learn.
- Studio UI is the local interactive surface for ZIP import or local-path analysis, run orchestration, and artifact navigation.
- Studio reuses the existing scanning/run/artifact pipeline and does not change Fact Engine semantics.
- Studio is not a hosted cloud platform by default.
- Facts:
- Truth-bearing run outputs from analysis.
- Example:
report.json,event_graph.json,api_surface.json,quality_gate.json.
- Patterns:
- Deterministic, evidence-linked engineering patterns derived from Facts.
- Example:
patterns.json,pattern_summary.json.
- Insights:
- User-facing derived outputs driven by Facts and Patterns.
- Example:
ai_summary.*,ai_risks/*,ai_explain/*,ai_ask/*.
- Primary truth source remains
detections.sqliteplus Evidence references. - Derived artifacts do not overwrite truth source.
run_manifest.jsonrecords artifact integrity and reproducibility metadata.
- Full-repo free roaming reduces determinism and auditability.
- Facts-first keeps cost and output drift under control.
- Source drill-down is available only on demand and bounded by Evidence refs + budget.
- Evidence-first
- Deterministic
- Facts first, source on demand