~/.khive/lore.db holds 341,967 full-content atoms (5.0GB — the expensive corpus), but the only access path is the separate mcp__lore__* MCP server. No lore CLI binary exists, so in any Bash context (subagents doing research, kkernel admin sessions) the entire library is unreachable — a big part of why the KHIVE.md-mandated lore compose pattern sees zero real usage.
Fix: register a lore. pack (suggest/compose/search/get_atom/stats) in the same verb registry kkernel and the khive MCP server share, so kkernel exec 'lore.suggest(query="...", role="implementer")' works everywhere other packs do. One DSL, all transports — consistent with the kkernel-as-admin-lane ruling.
~/.khive/lore.db holds 341,967 full-content atoms (5.0GB — the expensive corpus), but the only access path is the separate mcp__lore__* MCP server. No lore CLI binary exists, so in any Bash context (subagents doing research, kkernel admin sessions) the entire library is unreachable — a big part of why the KHIVE.md-mandated lore compose pattern sees zero real usage.
Fix: register a lore. pack (suggest/compose/search/get_atom/stats) in the same verb registry kkernel and the khive MCP server share, so kkernel exec 'lore.suggest(query="...", role="implementer")' works everywhere other packs do. One DSL, all transports — consistent with the kkernel-as-admin-lane ruling.