Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-09 · high
A paper and five open-source tools all reconstruct what an AI agent did or spent from console output, without added instrumentation.
Why now: As of September 9, a paper and five separate tools for parsing agent console output all existed side by side, covering both session timelines and spend.
Six small projects now watch AI coding agents work, without asking for any tracking code first.
Builders running Claude Code or Codex sessions have had no built-in way to see what an agent did, step by step. Tracking the cost meant wiring in logging by hand. These tools skip that step and read output the agent already writes.
Agent ATO reconstructs an agent's full interaction timeline straight from raw console output, described in a paper on reconstructing agent session timelines. It classifies each step into file discovery, reading, editing, or execution. No hooks. No SDK.
nickelsec/bough visualizes Claude Code sessions, prompts, and commits locally, so you can scroll back through a run after it finishes. tmux-agent-pulse puts live Claude Code and Codex status in a tmux status bar. It's built for people who work in a terminal and don't want to alt-tab to check whether a session stalled.
A separate group of tools tracks spending instead of actions. crwdla/tokentab, MeIotCOM/CodingPlanQuota, and codeburn each report cost across tools, reading the same kind of already-written output rather than requiring a new integration.
Each link below shares sources, entities, or timing with this story.
$3,054 against $38,370. Same benchmark, better score. Praxist (arXiv 2608.25955, submitted August 26) replaces per-attempt agent memory with a typed evidence graph of findings, plus lane-structured frontiers and agendas, so later attempts inherit validated mechanisms rather th...
QM went up under MIT license. Created July 29. As of the GitHub API check: 8,420 stars, 887 forks. Five days. YC uses it internally across accounting, legal, events, and engineering, including to build QM itself. Every employee and every Slack room gets its own scoped memory,...
I've spent real hours tuning the CLAUDE.md in my own repos. Rewriting architecture notes. Adding conventions. Trimming when it got long. So this one stung. arXiv 2607.27250 ran a two-agent ablation across Claude Code and Codex: 17 real tasks from 3 repositories, 288 gold-test-...
I check Product Hunt maybe once a week and usually regret it. Today's board is worth reading as market structure. The July 30 leaderboard: SKI at 277 upvotes (free voice input for Claude Code and Codex). AI Search Console at 249 (prompt analytics and citation mapping). Memmy A...
Beyond Yegge's Gas Town, Superset ("run an army of Claude Code, Codex"), agent-of-empires, and LobeHub's "Chief Agent Operator" all target managing many concurrent agents across harnesses, increasingly from mobile. The emerging UX problem is monitoring and steering dozens of p...
Microsoft's July 23 release targets a genuine gap: harness-based agents like Claude Code and Codex drive multi-turn reasoning, tool use, and external system access but were hard to train end-to-end with standard open RL infrastructure. The trick is decoupling training from inf...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.