EPIC: SIN-Brain memory cortex + automatic-calls rollout

[Delqhi]

Master plan: docs/plans/sin-brain-memory-cortex.md
PR (plans): #13

Goal: replace weak/passive memory + fragmented MCP tooling so OpenCode/Codex/Hermes agents share ONE memory cortex, a minimal whitelisted MCP surface, ONE AGENTS.md, and the same quality gates - with automatic calls (start-hook inject, agent-initiated tools, sleep-time reflect).

Role separation (binding): READ=SCKG, WRITE=Simone-MCP (kept, the moat), REMEMBER=SIN-Brain, VERIFY=Oracle/POC, door=sin serve.
Replace ONLY Serena (memory -> SIN-Brain, symbol-tools -> Simone). MCP whitelist = sin + simone.

SIN-Brain (new repo) workstreams:

• OpenSIN-Code/SIN-Brain#1 SB-1 storage + tiers
• OpenSIN-Code/SIN-Brain#2 SB-2 evidence graph
• OpenSIN-Code/SIN-Brain#3 SB-3 MCP memory tools
• OpenSIN-Code/SIN-Brain#4 SB-4 inject start-hook
• OpenSIN-Code/SIN-Brain#5 SB-5 reflect sleep-time
• OpenSIN-Code/SIN-Brain#6 SB-6 CLI + adapters

Bundle integration:

• BR-1: Register SIN-Brain memory tools in sin serve #14 BR-1 register memory tools in sin serve
• BR-2: sin agents-md embeds sin-brain inject block #15 BR-2 agents-md inject block
• BR-3: sin mcp-config <opencode|codex|hermes> generator #16 BR-3 mcp-config generator
• BR-4: OpenCode automatic calls via .opencode hooks #17 BR-4 OpenCode automatic-call hooks
• BR-5: Replace Serena fully (memory -> SIN-Brain, symbol-tools -> Simone) #18 BR-5 replace Serena (memory -> SIN-Brain, symbol-tools -> Simone)
• BR-6: Simone-MCP as canonical WRITE/edit layer (keep, do not replace) #20 BR-6 Simone-MCP as canonical WRITE layer (keep, do not replace)

Sequence: SB-1 -> SB-2/SB-3/SB-4 -> SB-5 -> SB-6 -> BR-1 -> BR-3 -> BR-6 -> BR-2 -> BR-4 -> BR-5.