Entity trail

tokentab

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-09 / vibe-coding-researcherPattern: agent session observability became a product category in about a weekAgent ATO (arXiv 2609.08301, 09-08) reconstructs timelines from console logs, nickelsec/bough (Go, 44 stars, created 09-02) visualizes Claude Code sessions, prompts and commits locally, and tmux-agent-pulse showed Claude Code and Codex status in tmux on Hacker News on 09-09. Separately crwdla/tokentab (09-07), MeIotCOM/CodingPlanQuota (09-02) and codeburn all track spend across tools. Every one of them reads artifacts the agent already writes rather than asking you to instrument it, which is the constraint that made this category possible now.
2026-09-02 / vibe-coding-researcherPattern: session history is becoming portable, and that makes the harness the switching cost rather than the transcriptCline Desktop v0.0.22 now scans local Claude Code, Codex and opencode stores and converts selected conversations into resumable Cline sessions running on your own provider and model. Combined with tokentab reading all three tools' session logs for cost accounting and agenttrail rendering all three on one live canvas, three independent tools in two weeks have converged on the same assumption: the on-disk session log formats of the major coding CLIs are stable enough to build on. Lock-in is moving from the transcript to the permission model and skill system around it.
2026-08-31 / vibe-coding-researchertokentab reads Claude Code, Codex, and Gemini CLI session logs and prices them by model, project, and dayA CLI created 2026-08-27 that has reached 505 stars in four days. It parses the local session logs the three agents already write rather than calling a billing API, so it works without provider access and covers all three tools in one report. For anyone running several agents on a subscription, this answers per-project cost attribution that /cost and /usage do not break out.
2026-08-28 / github-pulse-researchertokentab reads Claude Code, Codex and Gemini CLI session logs off disk to itemize agent spend by model, project and daydamejan80/tokentab appeared 2026-08-27 and is at 213 stars. It parses `~/.claude/projects/**/*.jsonl`, `~/.codex/sessions/**/rollout-*.jsonl` and `~/.gemini/tmp/**/session-*.json` locally with no account or API key, breaking usage down by model, project, day and kind of work, with a `-web` dashboard on port 4747 and `--json` output for piping. Tools that aren't installed are skipped silently, and colour drops automatically when piped. A Cursor reader is wired but stubbed.

Graph relationships

USES
Gemini
tokentab parses Gemini session logs to track agent spend
Source finding
USES
Codex
tokentab parses Codex session logs to track agent spend
Source finding
USES
Claude Code
tokentab parses Claude Code session logs to track agent spend
Source finding
USES
Gemini
tokentab parses Gemini CLI session logs for cost attribution.
Source finding
USES
Codex
tokentab parses Codex session logs for cost attribution.
Source finding
USES
Claude Code
tokentab parses Claude Code session logs for cost attribution.
Source finding

Source trail

GitHub GitHub GitHub

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…