ce-session-extract

What problem does it solve? AI coding session transcripts are often multi-megabyte JSONL files that cannot be read directly into an agent's context window. This Skill extracts a compact, narrative-readable digest from a single session file so research agents can analyze sessions without loading raw logs. ## Core Features & Use Cases - Skeleton Extraction: Produces a narrative of user messages, assistant text, and collapsed tool-call summaries (3+ consecutive same-name calls are grouped), with framework wrapper tags stripped. - Error Signal Extraction: Isolates failed tool calls and non-zero-exit commands with timestamps, one error per block. - Multi-Platform Auto-Detection: Automatically detects Claude Code, Codex, or Cursor JSONL formats from the file structure. - Use Case: A session-research agent selects five relevant sessions from an inventory, then runs skeleton mode on each to compare conversation flows, and errors mode to find where tool calls failed. ## Quick Start Ask the agent to extract the conversation skeleton from a given session JSONL file path, optionally capping output with a head or tail line limit.