v0.9.7
This release introduces a major improvement in the session Stats view. Transcript-based token usage analytics with per-model breakdowns, subagent tracking, and canonical pricing. Requires the server to have read access to claude and codex transcript files. Now enabled by default.
Features
- New Token Usage section in the Stats tab with per-model summaries, cost breakdowns, and fallback to events when transcripts are disabled
- New prompts token usage in session Stats - clickable link auto scrolls to the prompt event
- Cost breakdown tooltips on prompts and Est Cost cells with improved tooltip layout
- Longest tool call now renders as a clickable link to its PreToolUse event
- Codex transcript support with expanded test coverage
- Event deduplication at
/eventsingestion via canonical-JSON signature hashing and a UNIQUEsignature_hashcolumn AskUserQuestionevent details now render the question, options, and answer- New
/observe viewand/observe statsskill commands open the browser to the current session AGENTS_OBSERVE_TRANSCRIPT_STATSflag enabled by default, with Docker bind mount for~/.claude/projectsand~/.codex/sessions- "All" filter exclusions: seed default-all filter hides
PostToolBatch, gates Claude Code and default-agent events, and backfills missing seed filters on existing installs
Fixes
- Filter out subagents with zero LLM activity from stats
- Include subagents without transcript JSONLs as stub rows
- Miscellaneous Stats tab revamp fixes
Other
- Documentation: added session token usage image to README, plus design specs and implementation plans for transcript token stats, event deduplication, UI revamp, and All filter exclusions
- Chore: simplified Codex hooks env, removed old Claude Observe references, added transcript-stats setting to Claude config, applied formatter