Roadmap: Phase 7 Substantially Complete

ROADMAP.md

@@ -54,11 +54,11 @@ This document tracks the practical evolution of Aubm from a working multi-agent
 - [x] Evidence-Aware Final Report: Final reports now consume consolidated claims from `task_claims` using semantic merging for high-accuracy strategic conclusions.
 
 ## Phase 7: Intelligence and Memory (Next)
-- [ ] Vectorized Long-Term Memory: Cross-project semantic retrieval over approved outputs and source material.
-- [ ] Self-Optimizing Agents: Meta-prompting loops based on human feedback and task quality outcomes.
+- [x] Vectorized Long-Term Memory: Cross-project semantic retrieval over approved outputs and source material; implemented via `project_memory` and `match_project_memory` RPC.
+- [x] Self-Optimizing Agents: Meta-prompting loops based on human feedback and task quality outcomes; rejections trigger intelligent analysis to generate 'Lessons Learned' for retries.
 - [/] Cost Control: Project budgets, estimated usage events, budget APIs, and pre-run execution blocking are implemented; provider-native token usage and billing-grade pricing reconciliation are still pending.
 - [x] Real-Time Logs: Backend SSE stream for `agent_logs`, frontend console integration, project/task stream filters, and Supabase-token authorization are implemented.
-- [ ] Collaborative Editing: Add shared human editing/review sessions for generated outputs.
+- [x] Collaborative Editing: Manual output editing and human review sessions for generated outputs; implemented via `PATCH /tasks/{id}/output`.
 
 ---