provider-transcript-normalizerlisted

# Provider Transcript Normalizer Use this when adding or fixing RightMemory support for a provider's chat history files, such as Codex, Claude, Cursor, or another coding agent. ## Goal Do the one-time integration work so RightMemory can later scan that provider automatically. This skill is about reading provider transcripts and producing a small common session shape. It is not about deciding what should be remembered. ## Normalized Session Shape Normalize one reviewable session at a time: ```json { "source": "codex", "session_id": "abc", "project": "/path/or/null", "started_at": "2026-05-16T05:17:53Z", "ended_at": "2026-05-16T06:02:10Z", "turns": [ { "user": "user-visible message", "assistant": "assistant-visible final reply" } ] } ``` Keep this shape provider-neutral. Do not include tool calls, hidden reasoning, raw refs, streaming deltas, large metadata, or provider-specific event objects. ## Procedure 1. Locate the provider's transcript storage. 2. Inspect a small raw sample to identify sessions, timestamps, project path, user messages, assistant replies, and completion status. 3. Implement or update a provider adapter that emits normalized sessions. 4. Add source configuration so RightMemory can scan the provider later without the main agent calling it. 5. Add small fixture transcripts and tests for the adapter. 6. Verify repeated scans are deterministic and reviewed sessions are not processed again. ## Boundaries - Do no