// category
Terminal-native coding agents — Claude Code, Codex CLI, Aider, OpenCode, Gemini CLI.
CLI agents are terminal-native coding assistants — Claude Code, Codex CLI, Aider, OpenCode, Gemini CLI. Unlike IDE extensions, they live in your shell, read your whole repo, run tests, and open PRs. This category covers releases, workflow changes, and the migration from autocomplete-style AI to task-level delegation.
A developer series analyzing Claude Code's source code covers context management in its fourth installment, explaining that the coding agent rebuilds its full model request each turn from system rules, tool descriptions, message history, and compressed summaries, since the underlying model is sta...
A source code analysis of Claude Code describes its tool system architecture, in which a `Tool.ts` contract requires each tool to declare parameters, permissions, read-only status, and concurrency before the runtime executes any model-requested action.
Simon Willison published a piece arguing that HTML is a highly effective output format when using Claude Code, Anthropic's CLI coding tool, with accompanying examples hosted on GitHub Pages.
Anthropic has rented 300 megawatts of compute capacity from xAI's Colossus 1 datacenter, nearly 70% of its total capacity, as Claude usage grew 80-fold annually in Q1 2026. The deal doubled the 5-hour usage limit in Claude Code and expanded API rate limits, with Tier 1 output tokens rising from 8...
A developer published CodeSpec, a Claude Code plugin for spec-driven development workflows, on GitHub. The plugin persists state across six stages—spec, clarify, plan, tasks, implement, and review—allowing sessions to resume after interruptions without reloading full context.
Anthropic's Claude Code platform added a feature called Agent Skills — modular capability packs defined by a directory containing a SKILL.md file — enabling Claude to load context and scripts on demand rather than at startup. Claude Code runs on terminal, IDE, desktop, web, iOS, and Slack, using ...
OpenAI released a Chrome extension for its Codex product that allows agents to operate within a user's existing browser session, accessing signed-in sites, cookies, and authenticated workflows across multiple tabs. The extension connects Chrome to the Codex desktop app on Windows and macOS, enabl...
A developer released "cadenza," an open-source Claude Code plugin that structures technical writing into five sequential phases — issue finding, decomposition, storyboarding, verification, and output generation — with gate checks that prevent skipping steps. The plugin outputs a Markdown file and...
Amp, an AI coding startup that spun out of Sourcegraph in late 2025, released a rebuilt CLI called Neo, redesigned to support remote control of terminal sessions via a web interface, plugins, and longer-running agent workflows. Developers can start a local CLI session and manage it remotely, incl...
A blog post compares Anthropic's Claude Code CLI tool, priced at $20–200/month, against the cost of hiring a software developer at approximately $80,000 per year, concluding the tool functions as a developer aid rather than a full replacement.
OpenAI's "Codex for (almost) everything" update added an in-app browser, computer use, PR review, SSH connections, and over 90 plugins to its desktop app, used by more than 3 million developers weekly. In testing against the HTTPie Python codebase, Codex read a GitHub issue, traced a bug to three...
A developer released ai-config-sync-manager v0.1.0, a Node.js CLI tool that synchronizes configuration files between Claude Code and OpenAI Codex, translating between their differing formats for agents, permissions, and MCP servers. The tool runs via npx and supports six sync areas with automatic...
Spotify released a command-line tool called "Save to Spotify" that lets AI agents such as Claude Code and OpenAI Codex upload AI-generated audio summaries and podcasts directly to a user's Spotify podcast feed. The tool is available on GitHub and is triggered by adding "save to Spotify" to an AI ...
Anthropic held a "Code w/ Claude 2026" developer event featuring morning keynote sessions focused on Claude Code, its AI coding tool. Simon Willison live-blogged the event for simonwillison.net.
A developer published a tutorial for building a custom statusline in Claude Code using a shell script and jq, replacing the default display with fields showing context window usage percentage, 5-hour and 7-day rate limit consumption, and a terminal bell notification via a Stop hook in ~/.claude/s...
A Chroma 2025 study of 18 frontier AI models, including Claude 4, GPT-4.1, and Gemini 2.5, found all performed worse as input length increased, with some dropping from 95% to 60% accuracy past a context saturation threshold. The effect, called "context rot," is more pronounced in coding agents be...
Claude Code supports four hook handler types — command, prompt, agent, and http — across 21 lifecycle events. Command hooks run in under 5ms and produce deterministic results, while prompt hooks invoke an LLM and take 300–2000ms, and agent hooks spawn full Claude Code sessions with file and tool ...
A developer guide argues that OpenAI's Codex, an autonomous coding agent that reads repos and runs commands, performs better when given bounded "atomic" tasks with defined outcomes and verification steps rather than the open-ended conversational prompts suited to ChatGPT.
A benchmark across 12 production monorepos (4.2M lines of code) found Claude Code 2026 reviewed TypeScript PRs 45% faster than Codeium 2.0 (12.4s vs 22.6s), while Codeium 2.0 was 22% faster for Java/Kotlin repos; Claude Code 2026 costs $149/seat vs $109 for Codeium 2.0.
Claude Code CLI and Claude Desktop can be configured to use Amazon Bedrock as the inference backend by setting environment variables in ~/.claude/settings.json and providing AWS IAM credentials, removing the need for a separate Anthropic API subscription.
A developer released claudestat, an npm package that provides a real-time monitoring dashboard for Claude Code sessions, tracking token costs, tool calls, context usage, and detecting loops. It is installable via `npm i -g @deibygs/claudestat`.
A developer outlined six Claude Code workflow configurations for indie hackers: a pre-deploy shipping checklist, launch thread writer, support reply drafter, pricing page generator, architecture decision recorder, and competitor analysis tool. Each is designed to read actual codebase files and ou...
A developer documented their shift from Claude Code to OpenAI's Codex for personal projects, citing lower cost and improved model quality around GPT-5.4 and later versions. The guide outlines differences in tooling, subscription plans, and CLI vs. desktop app usage for Claude Code users evaluatin...
DeepClaude is an open-source project that combines Anthropic's Claude Code agent loop with DeepSeek V3/V4 Pro, routing reasoning tasks through DeepSeek's model while using Claude for code execution and tool use.
A developer compared Claude Code and Cursor over a 14-day sprint building 4 iOS apps across 11 repositories, finding Claude Code better suited for multi-repo and long-running tasks, while Cursor was faster to set up and better for React/Next.js frontend work.
A developer used Claude Code to build a complete notification system (schema, API, queue worker, tests) for a Next.js/Postgres project over one week without writing code manually. A missing `await` keyword on an async call in the generated worker code caused random 4-second notification delays in...
Simon Willison added a "Sightings" section to his blog that pulls in wildlife photos from his iNaturalist account, back-populating over a decade of observations totaling 208 entries. He built the feature using Claude Code on his phone as an extension of his existing content syndication system.
Claude Code loses all session context on restart; a developer workflow using a MEMORY.md file in the project root provides a 200-line persistent index of evolving project state, such as recent migrations and active decisions, which Claude Code reads at session start.
A developer released Scout, an MIT-licensed self-hosted research agent that uses Claude Code to convert GitHub Issues into cited research reports published to GitHub Pages. The system includes a pre-research "sharpening" step that clarifies ambiguous queries before dispatching parallel sub-agents...
A developer released Governor, an open-source Claude Code plugin that compresses memory files, filters build/test log output, and adds usage telemetry to reduce token consumption. Small local benchmarks reported approximately 55% reduction in output tokens and 96% of noisy pytest output blocked.
A Dev.to comparison of Claude Code 3.5 and Tabnine 2026 tested 50 developers on three coding tasks, reporting Claude completed work 35% faster on average, while Tabnine had lower completion latency (90ms vs. 120ms) and a 12% lower initial error rate.
Uber depleted its entire 2026 AI budget within four months by spending it on Claude Code, Anthropic's AI coding tool. The rapid spending indicates unexpectedly high adoption or usage costs among Uber's engineering teams.
Developer Robert Nowell released "skill-tree," a tool that analyzes Claude Code session history against 11 collaboration behaviors from Anthropic's February 2026 AI Fluency Index study, scoring users and assigning one of seven archetype cards. It is available as a Claude Code plugin and as an npm...
OpenAI released Codex CLI version 0.128.0, adding a `/goal` command that causes the coding agent to loop repeatedly until it determines a user-set goal has been completed or a configured token budget is exhausted. The feature is implemented via two prompt templates injected automatically at the e...
Anthropic released Claude Security, a codebase vulnerability scanning tool within Claude Code, from closed preview to public beta for Enterprise customers on Thursday. The tool uses multiple parallel agents to analyze data flows and includes a self-validation pipeline to reduce false positives, w...
A benchmark test on a 10,000-line Go 1.24 codebase with 412 injected bugs found Claude Code 2.5 detected 89.3% of defects versus 76.1% for Codeium 1.8, though Codeium processed files 50% faster and costs $20 less per seat monthly.
A developer tutorial describes wrapping Anthropic's Claude Code CLI in a bash loop to create a persistent polling agent, avoiding per-tick cold-start costs from MCP server handshakes and file re-reads that can add several seconds per interval.
A technical comparison of four AI coding tools finds Claude Code and Gemini CLI operate as terminal agents, Cursor integrates with VS Code, and Codex focuses on automated task execution. All four store session history as JSON files with no cross-project search or cross-session memory.
Claude Code stores conversation history as JSON files in ~/.claude/projects/ (macOS/Linux) or %USERPROFILE%\.claude\projects\ (Windows), with filenames based on hash strings that carry no semantic information. The tool lacks a native interface for browsing or searching past sessions across projects.
A hallucinated Terraform configuration generated by Claude Code 3.5 caused a SaaS startup to incur $51,237 in AWS charges over 72 hours after the AI incorrectly specified 120 m5.24xlarge instances instead of 2 m5.large instances for an EKS node group.
A tutorial describes using Claude Code's Stop hook and PostToolUse hook to build a three-layer verification loop (syntax, intent, regression) that prevents the agent from completing until checks pass. The approach references a 13.7-point benchmark gain from LangChain's similar PreCompletionCheckl...
A software engineer with a background in trading systems at Bloomberg described shifting from skepticism about LLMs to regular use of Claude Code after finding that structuring interactions with proper context improved results more than prompt wording alone.
SonarQube's static analysis tools can be integrated into Claude Code via a three-layer stack comprising sonarqube-agent-plugins, sonarqube-cli, and a containerized sonarqube-mcp-server. The integration requires SonarQube Server 10.x or later, as the MCP server calls the /api/v2/ endpoints not ava...
In Claude Code, token consumption is dominated by system prompts, conversation history, tool definitions, and CLAUDE.md project files rather than the user's typed input. Anthropic recommends keeping CLAUDE.md under 200 lines and notes that HTML block comments in that file are stripped before cont...
A developer described a method to connect OpenClaw, an open-source AI agent framework, to Claude by using the authentication built into the Claude Code CLI binary, bypassing the need for a separate API key or web account.
A developer built `claude-recall`, a tool that indexes Claude Code's JSONL session archives into SQLite with FTS5 full-text search and injects relevant prior sessions into new prompts via a `UserPromptSubmit` hook. The tool optionally uses a local Ollama embedding model for semantic reranking, wi...
A developer published a three-stage workflow for using Anthropic's Claude Code that requires AI to first produce written research and implementation plans in Markdown files before generating any code. The approach separates analysis, planning, and execution to reduce unreviewed code output.
Anthropic confirmed that user complaints about degraded Claude Code performance over the past two months were caused by three separate bugs in the Claude Code harness, not the underlying models. One bug, introduced March 26, caused session memory to be cleared every turn after an idle period rath...
OpenAI published a guide for getting started with Codex, its AI-based coding agent, covering project setup, thread creation, and task completion. The guide is aimed at new users beginning to work with the tool.
A developer tutorial describes using Anthropic's Claude Code CLI to generate n8n workflow JSON files from natural language prompts, bypassing manual node configuration in n8n's visual editor. The approach involves prompting Claude to produce importable JSON that n8n can execute across its 400+ in...
OpenAI published documentation for configuring Codex settings, covering options for personalization, detail level, and permissions to customize how the AI coding agent runs tasks.
OpenAI added an Automations feature to Codex that allows users to schedule and trigger automated tasks, including report generation, summaries, and recurring workflows without manual intervention.
Anthropic published an engineering postmortem on April 23rd acknowledging quality regressions in Claude Code, its agentic coding product. The degradation occurred in the product's orchestration and prompting layer, not in the underlying Claude API, which remained unchanged.
OpenAI offers Codex, an agent-based product designed to automate tasks, integrate with external tools, and generate outputs such as documents and dashboards, extending beyond standard chat interactions.
OpenAI's Codex supports plugins and skills that allow users to connect external tools, access data sources, and define repeatable workflows to automate tasks.
A developer described building a system of 90 Claude Code agents organized across 7 departments with hierarchical routing, using CLAUDE.md context files to give each agent a defined role, data access, and escalation path. The setup replaces generic AI prompting with specialized agents invoked by ...
A developer who spent $800 on Claude API tokens over six months building with AI coding agents launched a visual management tool for Claude Code called MC-MONKEYS and plans to publish guides on AI agent workflows and token cost optimization.
A developer built GeoSim, a geopolitical simulation engine using Next.js 14, Supabase, and the Anthropic API, in which six AI agents representing world powers simultaneously plan moves across branching timelines. The project used Claude Code's hook system to auto-run tests on file save and a 216-...
A developer implemented the 1993 DOOM engine running inside Claude Code's terminal statusline, using the doomgeneric C library to render frames as 24-bit ANSI and exposing game controls via a UserPromptSubmit hook and MCP server. The project uses four of Claude Code's existing extension points an...
A solo developer ran a 30-day experiment requiring all backend code for a client's Node.js/PostgreSQL scheduling API to be drafted by Claude Code first. The experiment produced fast results on routine backend tasks but encountered a significant failure on day 15 involving a database error.
A developer published a method for adding persistent memory to Claude AI coding agents using a file-based system with a lightweight index loaded each session and on-demand retrieval of structured markdown files covering user profiles, project state, corrections, and external references. The appro...
Two students built Rally, a location-based social platform for posting and joining local activities, as a class project over two 2-week sprints using Anthropic's Claude Code for planning, coding, testing, and deployment. Their workflow included a CLAUDE.md configuration file, automated lint and t...
Whoff Agents, a set of Claude Code skill packs and tools, launched on Product Hunt on April 21st, offering products priced from $29 to $99, including a TDD/debugging skill pack, a Next.js SaaS boilerplate, and an MCP security scanner covering 22 attack vectors.
Using `git worktree`, developers can assign each of multiple parallel Claude Code instances its own isolated working directory and branch, preventing `git stash` operations in one instance from overwriting uncommitted changes in another. The approach replaces stash with WIP commits before rebasing.
A developer compared Claude Code and opencode over six weeks, finding Claude Code faster on a refactoring task (9 vs. 16 minutes) with more mature multi-step workflow tooling, while opencode supports 75+ AI providers, is free as a standalone tool, and allows local model execution.
A tutorial contrasts using Claude as a chat tool versus agent mode, where Claude Code and Model Context Protocol (MCP) allow the model to read codebases, edit files, run tests, and interact with external tools like GitHub and Slack autonomously. A cited survey found 55% of engineers regularly use...
A developer published a method using Claude Code CLI's Schedule feature to automate SaaS customer support, running hourly checks that classify tickets into three categories: FAQ auto-replies (similarity score above 0.7), automated bug fixes, or human escalation for billing and complex issues.
In Claude Code, system prompts are ephemeral API-level instructions that reset each session, while CLAUDE.md is a persistent, project-scoped file stored in the repository that Claude reads automatically at session start. When the two conflict, CLAUDE.md instructions are treated as high-priority p...
A developer released "claud-code-account-switcher," an npm package that allows Claude Code users to switch between multiple accounts while preserving each account's authentication, history, plugins, and MCP server configurations. It is available via `npm install -g claud-code-account-switcher`.
Simon Willison updated his blog-to-newsletter tool to include a new content type called "beats" — posts capturing external activity like open source releases and museum visits — by prompting Claude Code to clone a reference GitHub repo and modify the relevant HTML file in a single session.
OpenAI released a major update to Codex, used by over 3 million developers weekly, adding background computer use, an in-app browser, image generation via gpt-image-1.5, more than 90 new plugins, GitHub PR review support, SSH connectivity, scheduled task automations, and a memory feature for reta...
OpenAI updated its Codex agentic coding tool with expanded desktop control capabilities, positioning it as a competitor to Anthropic's Claude Code. The update gives Codex broader ability to interact with a user's desktop environment.
OpenAI updated its Codex desktop coding tool with the ability to operate desktop apps on macOS, generate images via gpt-image-1.5, browse the web natively, schedule tasks, and retain memory from past sessions. The update also adds plugins for GitLab, Atlassian Rovo, and Microsoft Suite, with EU a...
A developer released claude-studio, an open-source visual orchestration platform for managing multi-agent workflows using Anthropic's Claude Code. The tool provides a graphical interface for coordinating multiple Claude AI agents working in parallel.
Claude Code, Anthropic's command-line coding tool, received a new sidebar interface. A developer noted the visual update favorably but indicated concerns or caveats about it in a post on Dev.to.
Anthropic released a redesign of its Claude Code desktop app featuring a sidebar for multi-project session management, an integrated terminal pane, a side chat function (Ctrl+;) for context-aware queries, and consolidated model and effort controls.
Anthropic launched "routines" for Claude Code, allowing automated tasks to run on schedules, via API calls, or GitHub webhooks on Anthropic's cloud infrastructure, replacing manual GitHub Actions setups for tasks like issue triage and smoke testing.
Anthropic released a redesigned Claude Code desktop app with an integrated terminal, improved diff viewer, side chat functionality, and rearrangeable interface panes for managing multiple coding sessions simultaneously.
Anthropic released documentation for Claude Code Routines, a feature within its Claude coding platform available at code.claude.com.
Ten open-source GitHub repositories provide extensions and integrations for Claude Code, including Repomix for codebase context, Dify and Flowise for visual workflow builders, and Onyx for self-hosted AI alternatives. Installation is available via terminal commands or plugin marketplace.
Vercel added Sandbox management to its CLI tool through a new `vercel sandbox` subcommand, eliminating the need for a separate command-line tool. The feature is available in Vercel CLI version 50.42.0 and later.
Anthropic's February-March 2026 updates to Claude Code—including adaptive thinking, lowered default effort settings, and hidden reasoning display—contributed to degraded performance on complex engineering tasks, with the community identifying under-allocated reasoning budgets and system prompt bi...
Claude Code can be configured to use self-hosted models by setting ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN environment variables, then running claude with the --model flag to specify a local model like qwen3-coder-next.
MemCTX is an open-source MIT-licensed tool that maintains session memory for Claude Code by storing sessions in SQLite, auto-generating summaries via Claude API, and injecting relevant history into new sessions through a dashboard interface.
GitHub introduced Rubber Duck, an experimental feature in Copilot CLI that uses a second AI model to review coding agent plans before execution. Testing showed Claude Sonnet paired with GPT-5.4 as Rubber Duck achieved 74.7% of the performance gap between Sonnet and Opus, with larger gains on comp...
Users reported that Claude Code became unsuitable for complex engineering tasks following February updates, based on a GitHub issue that gained significant discussion on Hacker News.
Claude Code, an AI-powered code assistant, identified a Linux vulnerability that had remained undiscovered for 23 years.
Vercel announced its plugin now supports OpenAI Codex and the Codex CLI, providing access to 39 platform skills, three specialist agents, and real-time code validation for development workflows.