Tag: ai-agents
All the articles with the tag "ai-agents".
-
OpenAI Tripled Its ARC-AGI-3 Score by Fixing Its Own Plumbing
Retained reasoning and compaction took GPT-5.6 Sol from 13.3% to 38.3% on the ARC-AGI-3 public set with 6x fewer output tokens. The engineering lesson is solid. The 38.3% is not comparable to the 30.2% Opus 5 posted five days earlier.
-
OpenCode Denies `git`. It Runs `$(which git)` Just Fine.
A viral teardown of OpenCode, the 161k-star coding agent, shows its bash permission system parsing commands into a tree-sitter AST and still waving through env git, base64 pipes, and python subprocess. Textual command filtering is security theater, and the harness is the only trust boundary that matters.
-
Weekly Roundup: One Adverb, One Agent, and Twelve Hosted Runtimes
GitHub's injection scanner fell to the word 'Additionally', a single-agent red-teaming harness beat the swarm it shipped with, and the managed-agent market got mapped.
-
One Word Beat GitHub's Guardrail, and the Word Was 'Additionally'
Noma Labs' GitLost shows an unauthenticated attacker leaking a private repo's contents by opening one public GitHub Issue. GitHub built a scanner to stop exactly this; prefixing the malicious line with 'Additionally,' walked the data straight past it.