Coding Agents

Cursor says Composer 2.5 is 90% off in the SDK this weekend. Same-day posts also split between fast one-shot SDK builds and complaints about dropped connections, project controls, and agent follow-through compared with Codex, so test reliability before committing.

Anthropic previewed a Claude Code update that shows token usage by Skills, Agents, MCPs, and Plugins in the CLI, with desktop support planned later. The command targets skills-heavy workflows where token costs are hard to attribute across multiple moving parts.

Anthropic made /fast default to Opus 4.7, added prompt cache diagnostics in Claude Console, published scale guidance for large codebases, and renamed extra usage to usage credits. The bundle gives teams clearer controls over latency, cache misses, and paid usage while running long Claude Code sessions.

Anthropic said Claude now writes more than 90% of its code, while Alex Albert pushed lighter scaffolding, real-trace evals and self-check loops. Teams using Claude Design, Projects and skill descriptions should keep layout, routing and long-running agent work aligned.

Skilled launched a CLI and TUI that scans installed skills across Claude Code, Codex, Droid, OpenCode and Grok Build. It surfaces dead skills, single-project dependencies and usage by agent or project, so teams can clean up skill sprawl.

Posts said Codex usage limits were reset across paid plans as users shared Mac app feedback, browser control, and repo-review results. The examples show Codex being used as a daily driver for debugging and code audit work, so watch the limits if you rely on it for regular use.

Anthropic's ClaudeDevs account said it reset everyone's 5-hour and weekly rate limits. The reset landed alongside paid-user complaints about slow sessions and visible switching pressure toward Codex, without a root-cause or permanent policy change attached.

ChatGPT opened a Codex mobile preview that lets users review outputs, approve commands, inspect diffs, and steer long-running agent work from a phone. It matters because Codex jobs no longer stay desktop-bound, though early users say the flow can still depend on a battery-draining host machine and a clunky app UI.

xAI opened an early Grok Build beta for SuperGrok Heavy users, and early testers surfaced /loop, /imagine, best-of-n, self-checking, and memory features. It matters because xAI is moving from chat into a coding-and-automation CLI surface aimed at shipping apps and workflows.

Anthropic added /goal to Claude Code for completion-checked runs, alongside /loop, /schedule, stop hooks, auto mode guidance, and an Opus 4.7 fast mode preview. Use /goal when a session needs to keep working until a defined condition is met; fast mode is opt-in now and becomes the default on Thursday.

Anthropic opened Agent View as a research preview, giving Claude Code one control pane for parallel sessions, skills dispatch, and quick replies. The change makes multi-session supervision a native workflow instead of a terminal-tab workaround.

New Hermes Agent and Claude Code playbooks mapped memory, skills, soul, crons, and nightly GitHub sync into repeatable personal-OS setups. The guides push agent workflows into daily content and admin tasks while surfacing security and stale-memory failure modes.

UI-TARS resurfaced as an open-source desktop-control stack while Opendesk described using accessibility APIs and marked elements instead of raw pixel guesses. The approach makes computer-use workflows more repeatable, but it still depends on human-oriented interfaces.

Creators reported Claude Code sessions hanging for minutes with no status feedback, and an Anthropic engineer said responsiveness improvements and self-serve debug logs are on the way. Users also say Claude Desktop now shows context-window usage, giving long sessions a clearer limit indicator.

Anthropic says this week's Claude Code release fixes long-running session bugs across 1M-context prompts, caching, auth fallbacks, MCP retries, and terminal rendering. The update targets the stability problems that surface in longer agent runs and heavier IDE workflows.

At Code with Claude San Francisco, builders showed Claude Code running 21-agent app pipelines across Figma, Jira, Confluence, and TestFlight. Users should watch for reliability strain as posts and conference recaps tie recent slowdowns to Anthropic's reported 80x growth.

OpenAI Codex CLI v0.129.0 adds Vim mode, redesigned resume flows, stronger plugin management, and hook controls, while GOALS also reached the Linux app. The update makes long-running refactors and persistent task loops more structured across CLI and app use.

Builder demos show Claude Code being run as a structured team of specialist agents, including a 21-agent setup that Aakash Gupta says shipped from idea to App Store submission in 72 minutes. The workflow shifts the bottleneck from typing code to specs, review, and product judgment, making Claude Code look more like a product-build system than a code assistant.

SubQ launched a sub-quadratic sparse-attention model with a 12 million token context window and opened early access alongside SubQ Code. The company claims 52x faster 1M-token performance than FlashAttention and under 5% of Opus cost, putting long-context coding workflows into a new price and latency band.

Users report OpenAI increased Codex limits about 10x on the May 5 reset, with much longer /goal sessions and more computer-use demos. That should extend unattended runs for app migrations and visual prototyping.

Claude Code added phone push notifications for long-running tasks and published 50-plus stability fixes across recent CLI releases, including up to 67% faster /resume on large sessions. The update matters because mobile alerts, auth fixes and lower-memory startup remove several workflow interruptions from daily coding use.

Anthropic support docs now say Claude Pro users in Claude Code need extra usage to access Opus, with Sonnet 4.5 as the default. Separate user posts report mismatched receipts and an unverified $200 overage case, making spend harder to predict.

OpenClaw contributors posted a voice-persona feature and fresh performance numbers that cut first output from 1s to 43ms. Separate posts describe 300-user sandboxed deployments and stronger PR, CI, and testing workflows, pointing to team-scale use beyond hobby demos.

Anthropic refreshed Claude Code on web and mobile with a sessions sidebar, routines view, faster responses, and claude --teleport handoff to the CLI. Use it to start work on web or mobile and continue in a terminal with branch state intact.

Codex App Server added a Fedora RPM package for Linux installs as users pushed Codex into browser control, 3D-print setup, and rapid game prototypes. Watch for more repeatable desktop workflows as Codex moves beyond chat-only experiments.

OpenAI launched GPT-5.5 in ChatGPT and Codex for coding, computer use, docs, sheets, and longer tool-driven tasks. Early tests showed stronger games and frontend builds, while pricing jumped again and Opus 4.7 comparisons started immediately.

Anthropic published a post-mortem on Claude Code regressions, said the problems lived in the harness rather than the models or API, and reset subscriber usage limits after fixes. The update matters for long agent sessions because recent complaints centered on reliability, wasted effort, and broken trust.

Claude Code introduced /ultrareview in research preview, sending parallel bug-hunting agents to scan critical changes and return findings in the CLI or Desktop. That matters because Pro and Max users get three free runs through May 5, and analysis threads frame it as a lower-noise answer to conventional AI review false positives.

Users posted WOZCODE, a Claude Code plugin that swaps in smarter file and search tools, and reported 54% lower cost, 68% fewer turns, and faster completion on the same Opus 4.7 task. The benchmark is community-run, but it includes install steps and repeatable commands.

OpenAI updated Codex with Mac app control, background computer use, image tools, ongoing tasks, and 90+ plugins, while Remotion added a one-click skill. Agents can now work inside desktop creative apps and stacks without blocking the visible cursor.

Anthropic put Claude Code routines into research preview and rolled out a rebuilt desktop app the same day. The update moves Claude Code toward event-triggered agents and parallel task review, so teams can test workflow automation.