Summary

Wire the existing terraphim_agent_evolution crate into the ADF orchestrator reconciliation loop to give agents persistent memory of prior runs (solution lineage).

Approach

• Add HashMap<String, AgentEvolutionSystem> field to orchestrator
• On each output event: parse structured content (NDJSON for opencode, JSON for Claude), store as MemoryItem in agent's MemoryEvolution
• On agent exit: record success/failure as Lesson, call create_snapshot()
• On handoff: include evolution_snapshot_key so receiving agent can load prior state

Critical Files

• crates/terraphim_orchestrator/src/lib.rs -- add evolution system field, feed output events, snapshot on exit
• crates/terraphim_agent_evolution/src/evolution.rs -- AgentEvolutionSystem (MemoryEvolution, TasksEvolution, LessonsEvolution)
• crates/terraphim_orchestrator/src/handoff.rs -- add evolution_snapshot_key to HandoffContext
• New: crates/terraphim_orchestrator/src/output_parser.rs -- parse agent output formats

Acceptance Criteria

• After agent run, VersionedMemory snapshot exists in persistence
• LessonsState contains at least one lesson per run
• Handoff includes evolution key
• cargo test --workspace passes

Dependencies

• Blocked by: Gitea chore(deps)(deps): bump @fortawesome/fontawesome-free from 6.6.0 to 7.0.1 in /desktop #117 (Phase 0: ADF remediation)
• Related: Phase 2: Session persistence for Claude Code agents in ADF #639 (session persistence), Phase 4: Structured activity log for ADF #641 (activity log), Epic: Operational Skill Store -- from failure capture to procedural memory #692 (operational skill store)
• Part of: Agentic Knowledge Evolution plan (knowledge/external/ai-agents/avo-agentic-variation-operators-nvidia-2026.md)