You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Added
Memory observability for Claude Code auto memory. Claude Code writes its own persistent auto memory (per-repository markdown under ~/.claude/projects/<project>/memory/) through ordinary Read/Write/Edit tools. The hook now emits a child memory-operation span under the tool_execution span whenever such a tool targets a file inside that directory, using the OpenTelemetry gen_ai.memory.* conventions — so auto memory shows up on Latitude's Memory page with per-record change history and diffs. Write → upsert_memory, Edit/MultiEdit → update_memory, Read → search_memory; gen_ai.memory.store.id is the <project> slug and gen_ai.memory.record.id is the file path within the memory dir. Edit bodies are read from disk at hook time (the tool call carries only a diff); subagent auto memory is covered via the same path.
LATITUDE_CLAUDE_CODE_MEMORY (default 1) emits memory-operation spans; set it to 0 to disable them. LATITUDE_CLAUDE_CODE_MEMORY_CONTENT (default 1) includes record bodies; set it to 0 to emit structure and counts only. Bodies also honor LATITUDE_REDACT_ATTRIBUTES via the gen_ai.memory.records key.