What problem does it solve? AI coding agents leave behind session logs, but reconstructing what an agent actually did — which files it read, what it searched for, which commands it ran — requires manually digging through raw JSONL rollout files. This Skill turns those logs into a normalized, auditable evidence report. ## Core Features & Use Cases - Multi-runtime log discovery: Locates session logs from OpenCode exports, Codex CLI rollout files (~/.codex/sessions/), and Spinosa session metrics, with a defined priority order. - Normalized event extraction: Shell scripts convert Codex JSONL rollouts and OpenCode export JSON into a unified event schema, pairing tool calls with their outputs. - Structured analysis: Produces analysis JSON with operation summaries, per-file access with coverage depth estimates, search keyword frequencies, and chronological tool-call timelines. - Evidence report generation: Populates a Markdown report template covering files read, searches, edits, commands, token/cost estimates, and gaps. - Use Case: After an agent session behaves unexpectedly, ask for an audit of that session to see exactly which files were read, what grep patterns were run, and the full timeline of tool calls. ## Quick Start Ask the agent to collect evidence from the most recent Codex session and produce a report of what files it read and what commands it ran.
