A March 29 Hacker News post pushed a GitHub issue alleging that Claude Code was running `git fetch origin` plus `git reset --hard origin/main` every 600 seconds against a user repo. The root cause is still unresolved, but the report sharply reopens the repo-safety question for agentic coding tools.
A March 2026 Hacker News thread pushed Stanford SCS’s `jai` to 604 points and 313 comments. The tool aims to contain AI agents on Linux by keeping the current working directory writable while placing the rest of the home directory behind an overlay or hiding it entirely.
Cursor said on March 26, 2026 that real-time reinforcement learning lets it ship improved Composer checkpoints as often as every five hours. Cursor's research post says the loop trains on billions of production tokens from real user interactions, runs evals including CursorBench before deployment, and has already shown gains in edit persistence, dissatisfied follow-ups, and latency.
A Hacker News-favored essay looks back from ChatGPT's November 2022 launch to Claude Code, vibe coding, and local LLMs, arguing that AI's real value is useful but still harder to measure than the hype suggests.
GitHub said on March 28, 2026 that Copilot CLI can create a robust test suite from the terminal by combining plan mode, /fleet, and autopilot. The linked GitHub docs describe /fleet as parallel subagent execution and autopilot as autonomous multi-step completion, making the post a concrete example of multi-agent testing workflows in the CLI.
A Hacker News discussion around the `.claude` folder guide frames Claude Code configuration as versioned project infrastructure rather than repeated prompt setup. The breakdown of `CLAUDE.md`, rules, commands, skills, and agents shows how teams can standardize workflows, but it also creates a new governance layer for instructions.
OpenAI Devs said on March 26, 2026 that plugins are rolling out in Codex, letting the agent work with common tools such as Slack, Figma, Notion, and Gmail. OpenAI's Codex docs describe plugins as reusable bundles that package skills, app integrations, and MCP server settings, turning Codex into a more shareable workflow layer for teams.
A Show HN post introduced Cog, a file-based memory architecture for Claude Code that stores working memory, archival context, and self-maintenance workflows as markdown and git-visible conventions. Instead of adding a database or runtime, it treats the filesystem, CLAUDE.md, and scheduled skills as the interface for persistent agent memory.
OpenAI plans to acquire Astral, the Python tooling company behind uv, Ruff, and ty, subject to regulatory approval. The deal ties Codex’s rapid growth to a deeper push into the tools developers use before and after code generation.
Ollama said on March 26, 2026 that VS Code now integrates with Ollama via GitHub Copilot. Ollama docs say VS Code 1.113+, GitHub Copilot Chat 0.41.0+, and Ollama v0.18.3+ let users load local or cloud Ollama models into the Copilot model picker, with GitHub Copilot Free sufficient for custom model selection.
OpenAIDevs said on March 27, 2026 that Codex usage limits had been reset across plans so users could try newly launched plugins. OpenAI's Help Center says Codex is temporarily available on Free and Go, paid plans are getting 2x rate limits, and plugins package reusable workflows built from skills, app integrations, and MCP configurations.
GitHub now lets users mention <code>@copilot</code> in a pull request to request changes on that same PR. The company says Copilot coding agent handles the work in a cloud development environment, runs tests and linting, then pushes updates; pull requests from forks are not yet supported.