Agenttrace
agenttrace is a local-first TUI for auditing and debugging AI coding agent sessions. It reads agent session logs and surfaces per-session breakdowns of cost, token usage, wall-clock time, tool call failures, latency, file diffs, health summaries, and CI gate results. The interface is designed for post-session analysis: understanding what an agent did, where it failed, how long each step took, and what it cost.
The tool targets the observability gap in agentic development workflows where sessions can span dozens of tool calls and significant cost without any structured audit trail.
Placed in Tools > Assess at inner position because session-level observability for AI coding agents is a genuine need that becomes more acute as agent usage scales. agenttrace complements cost dashboards like codeburn (which tracks spend across sessions) by providing depth within a single session — failure analysis, latency breakdown, and diff-level detail. The local-first, log-reading approach means no instrumentation is required. Worth trying for anyone running long or expensive agent sessions who needs to understand and improve their agent workflows.
