feat: Memory approvals context links — file, terminal, and task references

[Abernaughty]

Summary

Enhance memory approval entries with contextual links — related files, terminal output, sandbox results, and task/agent context that produced each memory write.

Context

Per the Dashboard Workflow Roadmap, Section 6. Memory entries currently show content, tier, confidence, and trust assessment. This enhancement adds the "why" — linking each entry back to the context that generated it, enabling informed approve/reject decisions.

Design

Each memory approval entry displays:

• Related files with clickable links to file viewer
• Terminal output / sandbox results that triggered the memory write
• Task and agent context: which task, which agent, what step in the pipeline
• Trust assessment (existing): sandbox origin, confidence score, tier info

Data Flow

flush_memory_node already emits written_entries via the memory_writes_flushed GraphState field. This enhancement enriches those entries with:

• source_task_id — links to parent task
• source_agent_id — which agent generated this
• source_step — which pipeline step (e.g., developer, qa, sandbox_validate)
• related_files — files involved in the step that produced this memory
• source_output_ref — reference to sandbox result or tool call output

Acceptance Criteria

• Memory entries include source_task_id, source_agent_id, source_step metadata
• Related files displayed with clickable links
• Sandbox result / tool call output linked from memory entry
• Task context (objective, blueprint title) shown in memory detail view
• Enriched metadata passed through SSE to dashboard
• Existing trust assessment UI unchanged (additive enhancement)

Dependencies

None — builds on existing memory approval UI and flush_memory_node data.

Effort

Small (half session)

Source

Dashboard Workflow Roadmap, Section 6 — Memory Approvals Enhancement