Native Harness Memory Bridge

Signet provides one portable memory interface across harnesses. When a harness already has its own memory system, Signet indexes that harness’s native memory artifacts as external provenance-bearing artifacts rather than re-extracting, rewriting, or claiming ownership of those memories.

V1 implements the generic source abstraction for harness-owned memory artifacts. Codex memory files under ~/.codex/memories/ and automation-local memory files under ~/.codex/automations/*/memory.md are indexed into Signet’s artifact search layer and surfaced through recall as native_memory results. Claude Code’s native memdir files under ~/.claude/projects/*/memory/, ~/.claude/session-memory/, and ~/.claude/agent-memory/ use the same bridge path instead of the previous Claude-only watcher that rechunked MEMORY.md into Signet-authored rows. When native memory files disappear, Signet marks their artifact rows deleted and excludes them from active recall rather than hard-deleting provenance. Normal Signet hooks remain the write path for Signet memories created from Hermes, OpenClaw, OpenCode, Claude Code, and Codex sessions.

Codex’s MCP registration keeps hooks and non-duplicative Signet capabilities, but filters Signet memory tools by default so native memory plus hook-driven recall do not create two competing memory UXs.