Did Codex Reset
GitHub

Microsoft paper reports a 16.6-point average gain for CASD prompt optimization

DAIR.AI

DAIR.AI describes a Microsoft paper on coding agents as prompt optimizers. Its method, CASD, gives an off-the-shelf coding agent the full set of agent logs and has it write the analysis code, instead of running a search loop over small batches of trajectories. The agent computes statistics over the whole trajectory corpus, finds recurring failure modes, reads representative episodes, and writes those findings as rules in one prompt. It needs no environment access and no validation data.

Across ALFWorld, tau2-bench retail and telecom, and Spreadsheet Bench-Verified, one CASD pass improved the unoptimized baseline by 16.6 points on average. GEPA improved it by 10.9 points and SkillOpt by 5.3. Each optimized prompt costs about $1.60, more than 22 times cheaper than validation-gated search.

#prompt-optimization#casd#microsoft#coding-agents#benchmarks

Only 35.8% of the cleanest public terminal-agent RL collection passed a Salesforce AI Research audit

Salesforce AI Research put that collection at 35.8% clean and two others at 10.1% and 3.3%, after finding reward errors in both directions. The authors say RIVER, which filters defective environments, increases RL gains by 106% on Terminal-Bench-Lite and 30% on Terminal-Bench v2.1 while using fewer than 30% of TMax's environments.