AgentWatch crash logs for coding agents

Crash-style stack traces and replay for every AI coding agent action

Problem

Developers running long AI coding agent tasks often end up with a broken result and no idea which step went wrong. The agent silently edits the wrong file or retries a failing API, and there is no trace to inspect.

Solution

A lightweight SDK or wrapper that logs each tool call, file change, and reasoning summary into a searchable timeline. On failure, it replays up to the offending step and generates a shareable crash report.

Users

Individual developers and small teams heavily using coding agents like Claude Code, Codex, or Cursor

MVP

Intercept and log tool calls and file edits; save replayable per-session timelines; auto-capture N steps around a failure; local-first storage with no code upload; export Markdown report

Revenue

Free local tier for individuals; team subscription per seat for shared replays and reports; or metered by recorded sessions

Why now

Coding agents are moving from demos to daily long tasks, and observability is nearly empty; indie developers can claim this niche before big vendors standardize it