Ariana
Ariana is an observability and diagnostic tool for Claude Code agent sessions. It runs alongside Claude Code, captures a structured trace of each session — tool calls, agent reasoning steps, token consumption per action, and decision branch points — and provides a replay interface for analysing where a session succeeded or went off-track.
As Claude Code use extends beyond simple one-shot tasks to multi-step autonomous sessions, understanding session behaviour becomes an engineering concern. When an agent loops, makes wrong assumptions, or diverges from the intended direction, diagnosing the failure requires replaying the sequence of decisions — information that is not readily accessible from the terminal output alone. Ariana addresses this by functioning as a debugger for agent sessions rather than for code.
Placed in Assess at inner position because session observability is an emerging need that scales directly with the complexity and autonomy of Claude Code use. The inner position reflects that developers who run non-trivial Claude Code sessions should be evaluating Ariana now rather than relying on terminal scrollback alone. The tool also provides token usage breakdowns that are useful for cost management at scale.
The data captured by Ariana can inform improvements to CLAUDE.md rules and prompt structures, closing the loop between session failure and future session configuration.
