← All stories
● Covered by 6 sources · 37 reportsLow impact3 negative33 neutral1 positive

New Terminal Tools Manage AI Coding Agents and Sessions

🔄 Updated 2d ago — new reporting from The New Stack
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • Agent-Manager runs multiple AI coding agents in separate tmux sessions.
  • It offers a TUI for live status, prompting, and diffing agent changes.
  • Wallfacer manages and indexes AI coding session files from various agents.
  • Wallfacer allows naming, tagging, grouping, and searching of past sessions.
  • Both tools aim to improve developer workflow with AI coding assistants.
  • Proliferate is an open-source, self-hostable platform.
  • Proliferate runs Claude Code, Codex, OpenCode, Cursor, and Grok.
  • Proliferate provides isolated git worktrees for each task.
  • Proliferate supports subagents that delegate work to child agents.
  • Proliferate offers integrations for MCPs, skills, Computer Use, and Browser Use.
  • Proliferate includes workflows for recurring and event-driven agent runs.
  • Proliferate runs agents through their native harnesses.
  • Proliferate can be deployed with Docker, AWS, GCP, Azure, Kubernetes, and air-gapped operation.
  • Funes is a new tool providing a durable, local memory layer for coding agents.
  • Funes indexes agent session traces to allow agents to recall past decisions and rationales.
  • Funes supports Claude Code, Codex, pi, and Hermes agents.
  • Funes can travel to a Hugging Face dataset owned by the user, private by default.
  • Funes is a single binary with no ML runtime dependency for its default inference backend.
  • Claude Code runs on Firecracker microVMs.
  • Claude Code's VM uses a custom-built guest kernel (6.18.5-fc-v20).
  • Claude Code's VM init is a Rust binary named process_api.
  • process_api listens on vsock port 2024 to allow host control.
  • Geiger is a new open-source command-line tool.
  • Geiger inventories AI agents, harnesses, MCP servers, plugins, and extensions.
  • Geiger identifies what each AI component can access and execute.
  • Geiger is installed via `npx geiger-scan`.
  • Geiger does not require installation, an account, or telemetry.
  • Geiger reads configs and directories, and writes nothing by default.
  • Geiger can output to a JSON file.
  • An open-source agent harness gained 200,000+ GitHub stars in three weeks in August 2026.
  • The harness's plugin ecosystem passed 13,000 repositories in three weeks.
  • Geiger found 9 findings across 3 ecosystems on a machine.
  • Geiger reported 7 components could execute code and 1 credential was found.
  • OtoDock released a self-hosted platform for collaborative agents.
  • OtoDock allows agents to perform tasks, interact with tools, and work autonomously.
  • OtoDock is multi-tenant by design.
  • OtoDock offers four modes for sharing agents and their work.
  • OtoDock provides a self-hosted dashboard and phone line interaction for agents.
  • OtoDock agents run on user's Anthropic and OpenAI subscriptions, and local models.
  • OtoDock agents have a name and a job, like Personal Assistant or System Admin.
  • OtoDock agents are made of six editable parts.
  • AgentsDock is an IDE for agentic AI research.
  • AgentsDock supports macOS 14+.
  • AgentsDock allows researchers to manage multiple AI agents.
  • AgentsDock allows connecting to various servers.
  • AgentsDock allows accessing development tools from any device.
  • AgentsDock aims to centralize AI research workflows.
  • AgentsDock server is installed via git clone from ZhengyiLuo/AgentsServer.git.
  • AgentsDock allows side-by-side chat with different agents.
  • AgentsDock allows viewing and editing code on the remote.
  • AgentsDock provides full terminal access from anywhere.
  • Skillsync transfers AI chat sessions between different AI coding agents.
  • Skillsync converts proprietary session formats to prevent vendor lock-in.
  • Skillsync collects and makes all user sessions searchable.
  • Skillsync identifies loaded skills used by agents in sessions.
  • Skillsync allows creating shared workspaces for team session syncing.
  • Skillsync's core is an open-source Rust engine called txcript.
  • Claude Code's Projects feature allows managing multiple agents in a single cloud environment.
  • Claude Code Projects enable coordinated AI agent teams with shared memory and goals.
  • Claude Code Projects have 'threads' for parallel tasks and a 'coordinator' for overall progress.
  • Each thread in Claude Code Projects works on its own branch and repo copy.
  • Claude Code Projects resolve overlapping code work as merge conflicts.
  • Claude Code Projects will support local threads in the future.
  • Claude Code reads AGENTS.md if CLAUDE.md is absent.
  • Claude Code added CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1 for egress through a forward proxy.
  • Claude Code added an optional headers map on gateway upstreams for static headers to a proxy.
  • Claude Code fixed sessions hanging after internal errors.
  • Claude Code fixed conversations failing with "text content blocks must be non-empty".
  • Claude Code fixed unexpected logouts when an older build runs on the same machine.
  • A 'Chief of Staff' pattern separates orchestration from execution for long-horizon AI coding agents.
  • The 'Chief of Staff' pattern uses a coordinating session to manage tasks and verify outcomes.
  • The 'Chief of Staff' pattern stores state externally to overcome context limitations and unreliable self-reports.
  • Grok Build added memory on September 16.
  • Grok's memory uses Markdown files in workspace and global scopes.
  • Claude Code's auto memory keeps a MEMORY.md index plus one file per note, per repository.
  • Claude Code's auto memory is on by default.
  • Claude Code's Projects beta was announced September 17.
  • Claude Code's Projects beta is for select Pro and Max subscribers with no existing projects.
  • drawgent integrates AI coding agents with Excalidraw.
  • drawgent allows generating and modifying diagrams using natural language commands.
  • drawgent connects to Claude Code, Codex, or opencode installations.
  • drawgent uses a screenshot and scene to allow agents to look at the canvas.
  • drawgent requires Node.js >= 18 for Claude and Codex bridges.
  • drawgent setup command configures agents.
  • drawgent up command starts a new agent session and canvas.
  • drawgent up --attach connects to a running session.
  • drawgent writes configuration to ~/.config/drawgent/config.toml.
  • opencode speaks ACP itself.
  • Claude Code and Codex use official ACP adapters.
  • Offrun is a new workspace to manage multiple coding agents.
  • Offrun uses Git worktrees to isolate agent activities.
  • Offrun allows concurrent tasks like refactoring and bug fixes without file conflicts.
  • Offrun integrates peer review and account management features.
  • Offrun runs CLIs on the user's Mac, signed in as the user.
  • Offrun allows picking a reviewer whose findings land in chat while work is uncommitted.
  • Offrun allows pointing each agent at a different login.
  • Offrun moves chat to a login with room when one hits its limit.
  • Offrun keeps repo-wide context for project memory.
  • Berth is a new open-source tool.
  • Berth allows agents to run on remote development boxes.
  • Berth enables agents to continue working when a laptop is closed.
  • Berth is available for Apple silicon and Intel Macs.
  • Berth is signed and notarized.
  • Berth creates a worktree on a chosen box and starts an agent.
  • Berth displays agent commands, reads, and edits as diffs.
  • Berth allows comparing two worktrees in mirrored lanes.
  • Berth offers its own themes and nineteen ported themes.
  • Berth allows customizing chat background and text size.
  • Agent.reviews is a new platform where AI agents read and write reviews on development tools.
  • Agent.reviews provides setup steps and documentation for agents to interact with tools.
  • Agent.reviews lists Markdown versions of pages in llms.txt.
  • Claude Code API calls passed on the second try with one call per box.
  • A service manifest's request timeout was raised to 60 minutes.
  • uvx ran esptool and a pyserial + Pillow helper script without virtualenv or global install.
  • uvx repeat runs started fast from cache.
  • Terraform ran from its official container image on a new AWS configuration.
  • Terraform steps fmt -check, init -backend=false, validate, and providers lock passed on the first try.
  • Claude Code used dependency records from a build platform to understand an application's architecture.
  • The test aimed to provide context for AI agents beyond traditional memory functions.
  • The test aimed to help AI agents work on existing projects more effectively.
  • Bit Cloud is a platform that builds, hosts, and deploys applications from their first component.
  • Chainguard released two beta products: Workspaces and Checks.
  • Chainguard Workspaces provides remote development environments based on hardened microVMs.
  • Chainguard Checks moves continuous integration into the development loop.
  • Chainguard announced both products at its Assemble London event.
  • Workspaces gives each developer or agent a dedicated remote microVM.
  • The chainctl develop command synchronises a local working tree into the environment.
  • A Workspaces session is snapshotted when the user detaches and can resume.
  • Teams can create several isolated Workspaces for one repository.
  • Claude AI performs initial code reviews for the Claude Code team.
  • Claude Code team more than doubled merged pull requests between June and September.
  • Claude Code team reduced revert rate by over 40% between June and September.
  • Claude writes about 90% of Claude Code's own code.
  • Anthropic's monorepo changes per day increased from 2,500 in May to 30,000 in September.
  • Sphinx is a new open-source Git hook.
  • Sphinx quizzes developers on AI-generated code before it is pushed.
  • Sphinx is installed via `npm install -g sphinx-understand`.
  • Sphinx runs on minimal tokens from existing agents, OpenRouter, or OpenAI-compatible endpoints.
  • Sphinx works with Codex, Claude Code, Cursor, OpenCode, Pi, or plain Git.
  • Tmux provides persistent, scriptable sessions for long-running processes.
  • Tmux allows agents to survive disconnections, closed terminals, or dropped SSH connections.
  • Claude Code can create a worktree and tmux session with `claude --worktree --tmux`.
  • Tmux allows switching between multiple named agent sessions with a keystroke.
  • Tmux allows external scripts to type into or read from a running agent.
  • agent-usage-guard is a Claude Code mod.
  • agent-usage-guard prevents rapid consumption of the five-hour usage window.
  • agent-usage-guard monitors live usage and controls subagent fan-out, context churn, and retry loops.
  • agent-usage-guard acts before API calls are made.
  • agent-usage-guard prompts the user for approval or refuses actions to conserve usage.
  • agent-usage-guard reads the 5-hour and weekly percentages Claude Code reports.
  • Corgi is a new agent and terminal multiplexer.
  • Corgi tracks agent status: working, needs you, finished, idle, unread, clear.
  • Corgi allows jumping into work with Ctrl-T.
  • Corgi tracks token usage.
  • Corgi shows file changes made by agents.
  • Corgi organizes workspaces.
  • Corgi monitors resource usage per workspace live.
  • Corgi supports image pasting in the terminal, including over SSH.
  • Corgi requires Ghostty Nightly for image pasting.
  • Corgi allows undoing tab closes with Ctrl-Shift-T.
  • Corgi can be driven from any terminal using `corgi --skill`.
  • Claude Code sends each background agent's full transcript to its model every 30 seconds.
  • The transcript is sent to generate a 3-5 word status update.
  • This process consumes significant rate-limit headroom.
  • This behavior has been active since August 2026.
  • A JavaScript preload script can disable this behavior.
  • The anomaly was detected by DRIM in Claude Code's OpenTelemetry export.
  • agent_summary cost more than the main conversation itself.
  • Postman desktop application automatically installs AI plugins without user consent.
  • Postman installs Codex and Claude Code plugins.
  • Postman attempts to approve its own startup hooks.
  • Postman startup hooks route API engineering tasks through Postman skills.
  • Postman installs a user-wide plugin, a startup instruction hook, a remote MCP connection, a Claude Code plugin, and a Postman CLI.
  • The installed hook is named `codex-hooks.js`.
  • The Postman desktop application version 12.33.4 on macOS installed plugins.
  • The Postman installation event was marked auto-existing-upgrade.
  • Postman modified Codex's user-level configuration file, ~/.codex/config.toml.
  • Inkitt developed a system to automate small software changes from task intake to draft pull request.
  • Inkitt's system is triggered via Slack, Linear, or an API call.
  • Inkitt's system runs on remote infrastructure.
  • Inkitt's system ships code daily.
  • Inkitt's orchestrator persists each run in a database.
  • Inkitt's orchestrator separates decision-making from side effects.
  • Stroq is a new CLI tool that audits AI coding agent sessions.
  • Stroq identifies data leakage and security risks in agent activity.
  • Stroq analyzes commands executed and data sent by agents.
  • Stroq determines if sensitive information left the machine.
  • Stroq determines if commands repeated agent-read text.
  • Stroq reads Claude Code and Codex sessions.
  • Stroq shows how many commands came from text the agent read.
  • Stroq shows which keys left the machine.
  • Stroq reads the last 30 days of sessions, key files, and agent work folders.
  • Stroq prints key names but never values.
  • Stroq writes only a small cache under ~/.stroq.
  • Stroq sends no data, requires no account, and has no telemetry.
  • Stroq has commands like `init`, `why`, `attack`, `uninstall`, and `show`.
  • Stroq requires Node.js 22 or newer.
  • Stroq is licensed under Apache-2.0.
  • Stroq is a local security platform for AI coding agents.
  • Stroq's hooks judge every tool call before it runs.
  • Stroq denies `curl | sh` and outbound calls carrying known secrets.
  • Claude Code saves commands with secrets in .claude/settings.local.json when 'Yes, and don't ask again' is selected.
  • 1,122 public GitHub copies of .claude/settings.local.json contain at least one secret.
  • Removing .claude/settings.local.json does not remove secrets from git history or forks.
  • Apiiro reported the issue.
  • Atuin 18.24 introduces Atuin Threads.
  • Atuin Threads allows syncing, searching, and resuming AI agent sessions across multiple machines.
  • Atuin Threads enables transferring workflows between different agents and devices.
  • AgentRun is an open-source platform for AI agents.
  • AgentRun adds self-improvement, durable execution, and a control plane to agents.
  • AgentRun aims to reduce token usage and improve fault tolerance.
  • AgentRun agents learn from past runs.
  • AgentRun agents can resume execution after interruptions.
  • AgentRun provides a unified console for agent management.
  • AgentRun can tune a code reviewer on custom PRs for 18 cents a review.
  • AgentRun can classify merchants for 12 cents each.
  • Agweb is a new macOS browser for running multiple coding agents.
  • Agweb uses WebKit to prevent focus stealing and reduce resource consumption.
  • Agweb allows agents to be driven from the shell with the `agweb` command.
  • Agweb provides a live dashboard view of all agent activities.
  • Agweb suspends idle tabs, reducing CPU usage to 0%.
  • Agweb handles dialogs, changing tabs, file pickers, and downloads in code.
  • Agweb groups tabs by agent in the sidebar.
  • Agweb allows named profiles to share logins across agents.
  • Windows 11's MissionControlAggregator.dll records usage of developer tools and AI coding agents.
  • The recorded data is saved locally and reported to Windows Diagnostics.
  • The data is uploaded to Microsoft if users have not opted out of optional diagnostic data.
  • MissionControlAggregator.dll explicitly lists executable names like codex.exe, claude.exe, copilot.exe, cursor.exe, python.exe, dotnet.exe, gcc.exe, and code.exe.
  • Vercel launched Board, an experimental shared message board for AI agents.
  • Vercel used Board to coordinate security agents and rebuild v0's software factory.
  • v0's software factory now merges over 50% of pull requests.
  • IDrive launched AI Agents Backup, a feature in its desktop application.
  • IDrive AI Agents Backup backs up local data from Claude Code, Cursor, Codex, and Cline.

Overview of New AI Agent Management Tools

Developers now have access to two new terminal-based utilities designed to streamline the use of AI coding agents. Agent-Manager provides a terminal user interface (TUI) for running and monitoring multiple AI coding agents concurrently, while Wallfacer focuses on organizing and managing the session files generated by these agents.

Agent-Manager: Concurrent Agent Execution

Agent-Manager allows users to run AI coding agents such as Claude Code, Codex, OpenCode, and Grok side-by-side, each within its own tmux session. This setup ensures that agents continue to operate even after the manager is closed. The tool centralizes monitoring, displaying a live status list of all sessions, grouped into a project tree.

Users can send prompts directly to an agent's pane, spawn new agents, and revive dead sessions. Agent-Manager also includes a feature to open a full-file diff of changes made by an agent, with syntax highlighting, and allows comments on lines to be sent back to the agent.

Wallfacer: Organizing AI Coding Sessions

Wallfacer is a terminal session manager for AI coding agents like Claude Code, Cursor CLI, Kiro CLI, and Codex. It addresses the issue of scattered and unorganized session files by indexing them. Claude Code stores conversations as JSONL files, Kiro CLI uses a flat folder, Cursor CLI buries chats in hash-named directories, and Codex files rollouts by date.

Wallfacer provides a unified interface to name, tag, group, search, resume, or delete these sessions from a full-screen browser or the command line. It operates in a read-only capacity, never modifying the agents' original files.

Installation and Compatibility

Agent-Manager can be installed via `brew install yoanwai/tap/agent-manager` (which installs tmux if missing) or `go install github.com/YoanWai/agent-manager@latest` (requiring Go 1.26+ and tmux). It is also available for download from Releases for macOS and Linux (amd64/arm64). Status detection is supported for Claude Code, OpenCode, Codex, and Grok Build, with configuration options for other CLI tools.

Updates

🕒 2026-11-19 · new reporting from The New Stack
  • Vercel launched Board, an experimental shared message board for AI agents.
  • Vercel used Board to coordinate security agents and rebuild v0's software factory.
  • v0's software factory now merges over 50% of pull requests.
  • IDrive launched AI Agents Backup, a feature in its desktop application.
  • IDrive AI Agents Backup backs up local data from Claude Code, Cursor, Codex, and Cline.
🕒 2026-11-17 · new reporting from Hacker News Front Page
  • Windows 11's MissionControlAggregator.dll records usage of developer tools and AI coding agents.
  • The recorded data is saved locally and reported to Windows Diagnostics.
  • The data is uploaded to Microsoft if users have not opted out of optional diagnostic data.
  • MissionControlAggregator.dll explicitly lists executable names like codex.exe, claude.exe, copilot.exe, cursor.exe, python.exe, dotnet.exe, gcc.exe, and code.exe.
🕒 2026-11-14 · new reporting from Hacker News Front Page
  • Agweb is a new macOS browser for running multiple coding agents.
  • Agweb uses WebKit to prevent focus stealing and reduce resource consumption.
  • Agweb allows agents to be driven from the shell with the `agweb` command.
  • Agweb provides a live dashboard view of all agent activities.
  • Agweb suspends idle tabs, reducing CPU usage to 0%.
  • Agweb handles dialogs, changing tabs, file pickers, and downloads in code.
  • Agweb groups tabs by agent in the sidebar.
  • Agweb allows named profiles to share logins across agents.
🕒 2026-11-12 · new reporting from Hacker News Front Page
  • AgentRun is an open-source platform for AI agents.
  • AgentRun adds self-improvement, durable execution, and a control plane to agents.
  • AgentRun aims to reduce token usage and improve fault tolerance.
  • AgentRun agents learn from past runs.
  • AgentRun agents can resume execution after interruptions.
  • AgentRun provides a unified console for agent management.
  • AgentRun can tune a code reviewer on custom PRs for 18 cents a review.
  • AgentRun can classify merchants for 12 cents each.
🕒 2026-11-11 · new reporting from Hacker News Front Page
  • Atuin 18.24 introduces Atuin Threads.
  • Atuin Threads allows syncing, searching, and resuming AI agent sessions across multiple machines.
  • Atuin Threads enables transferring workflows between different agents and devices.
🕒 2026-11-07 · new reporting from The Hacker News
  • Claude Code saves commands with secrets in .claude/settings.local.json when 'Yes, and don't ask again' is selected.
  • 1,122 public GitHub copies of .claude/settings.local.json contain at least one secret.
  • Removing .claude/settings.local.json does not remove secrets from git history or forks.
  • Apiiro reported the issue.
🕒 2026-11-05 · new reporting from Hacker News Front Page
  • Stroq is a new CLI tool that audits AI coding agent sessions.
  • Stroq identifies data leakage and security risks in agent activity.
  • Stroq analyzes commands executed and data sent by agents.
  • Stroq determines if sensitive information left the machine.
  • Stroq determines if commands repeated agent-read text.
  • Stroq reads Claude Code and Codex sessions.
  • Stroq shows how many commands came from text the agent read.
  • Stroq shows which keys left the machine.
  • Stroq reads the last 30 days of sessions, key files, and agent work folders.
  • Stroq prints key names but never values.
  • Stroq writes only a small cache under ~/.stroq.
  • Stroq sends no data, requires no account, and has no telemetry.
  • Stroq has commands like `init`, `why`, `attack`, `uninstall`, and `show`.
  • Stroq requires Node.js 22 or newer.
  • Stroq is licensed under Apache-2.0.
  • Stroq is a local security platform for AI coding agents.
  • Stroq's hooks judge every tool call before it runs.
  • Stroq denies `curl | sh` and outbound calls carrying known secrets.
🕒 2026-11-04 · new reporting from InfoQ
  • Inkitt developed a system to automate small software changes from task intake to draft pull request.
  • Inkitt's system is triggered via Slack, Linear, or an API call.
  • Inkitt's system runs on remote infrastructure.
  • Inkitt's system ships code daily.
  • Inkitt's orchestrator persists each run in a database.
  • Inkitt's orchestrator separates decision-making from side effects.
🕒 2026-11-04 · new reporting from The New Stack
  • The Postman desktop application version 12.33.4 on macOS installed plugins.
  • The Postman installation event was marked auto-existing-upgrade.
  • Postman modified Codex's user-level configuration file, ~/.codex/config.toml.
🕒 2026-11-03 · new reporting from Hacker News Front Page
  • Postman desktop application automatically installs AI plugins without user consent.
  • Postman installs Codex and Claude Code plugins.
  • Postman attempts to approve its own startup hooks.
  • Postman startup hooks route API engineering tasks through Postman skills.
  • Postman installs a user-wide plugin, a startup instruction hook, a remote MCP connection, a Claude Code plugin, and a Postman CLI.
  • The installed hook is named `codex-hooks.js`.
🕒 2026-10-29 · new reporting from Hacker News Front Page
  • Claude Code sends each background agent's full transcript to its model every 30 seconds.
  • The transcript is sent to generate a 3-5 word status update.
  • This process consumes significant rate-limit headroom.
  • This behavior has been active since August 2026.
  • A JavaScript preload script can disable this behavior.
  • The anomaly was detected by DRIM in Claude Code's OpenTelemetry export.
  • agent_summary cost more than the main conversation itself.
🕒 2026-10-24 · new reporting from Hacker News Front Page
  • Corgi is a new agent and terminal multiplexer.
  • Corgi tracks agent status: working, needs you, finished, idle, unread, clear.
  • Corgi allows jumping into work with Ctrl-T.
  • Corgi tracks token usage.
  • Corgi shows file changes made by agents.
  • Corgi organizes workspaces.
  • Corgi monitors resource usage per workspace live.
  • Corgi supports image pasting in the terminal, including over SSH.
  • Corgi requires Ghostty Nightly for image pasting.
  • Corgi allows undoing tab closes with Ctrl-Shift-T.
  • Corgi can be driven from any terminal using `corgi --skill`.
🕒 2026-10-22 · new reporting from Hacker News Front Page
  • agent-usage-guard is a Claude Code mod.
  • agent-usage-guard prevents rapid consumption of the five-hour usage window.
  • agent-usage-guard monitors live usage and controls subagent fan-out, context churn, and retry loops.
  • agent-usage-guard acts before API calls are made.
  • agent-usage-guard prompts the user for approval or refuses actions to conserve usage.
  • agent-usage-guard reads the 5-hour and weekly percentages Claude Code reports.
🕒 2026-10-19 · new reporting from Hacker News Front Page
  • Tmux provides persistent, scriptable sessions for long-running processes.
  • Tmux allows agents to survive disconnections, closed terminals, or dropped SSH connections.
  • Claude Code can create a worktree and tmux session with `claude --worktree --tmux`.
  • Tmux allows switching between multiple named agent sessions with a keystroke.
  • Tmux allows external scripts to type into or read from a running agent.
🕒 2026-10-18 · new reporting from Hacker News Front Page
  • Sphinx is a new open-source Git hook.
  • Sphinx quizzes developers on AI-generated code before it is pushed.
  • Sphinx is installed via `npm install -g sphinx-understand`.
  • Sphinx runs on minimal tokens from existing agents, OpenRouter, or OpenAI-compatible endpoints.
  • Sphinx works with Codex, Claude Code, Cursor, OpenCode, Pi, or plain Git.
🕒 2026-10-16 · new reporting from Hacker News Front Page
  • Claude AI performs initial code reviews for the Claude Code team.
  • Claude Code team more than doubled merged pull requests between June and September.
  • Claude Code team reduced revert rate by over 40% between June and September.
  • Claude writes about 90% of Claude Code's own code.
  • Anthropic's monorepo changes per day increased from 2,500 in May to 30,000 in September.
🕒 2026-10-16 · new reporting from InfoQ
  • Chainguard released two beta products: Workspaces and Checks.
  • Chainguard Workspaces provides remote development environments based on hardened microVMs.
  • Chainguard Checks moves continuous integration into the development loop.
  • Chainguard announced both products at its Assemble London event.
  • Workspaces gives each developer or agent a dedicated remote microVM.
  • The chainctl develop command synchronises a local working tree into the environment.
  • A Workspaces session is snapshotted when the user detaches and can resume.
  • Teams can create several isolated Workspaces for one repository.
🕒 2026-10-11 · new reporting from The New Stack
  • Claude Code used dependency records from a build platform to understand an application's architecture.
  • The test aimed to provide context for AI agents beyond traditional memory functions.
  • The test aimed to help AI agents work on existing projects more effectively.
  • Bit Cloud is a platform that builds, hosts, and deploys applications from their first component.
🕒 2026-10-07 · new reporting from Hacker News Front Page
  • Agent.reviews is a new platform where AI agents read and write reviews on development tools.
  • Agent.reviews provides setup steps and documentation for agents to interact with tools.
  • Agent.reviews lists Markdown versions of pages in llms.txt.
  • Claude Code API calls passed on the second try with one call per box.
  • A service manifest's request timeout was raised to 60 minutes.
  • uvx ran esptool and a pyserial + Pillow helper script without virtualenv or global install.
  • uvx repeat runs started fast from cache.
  • Terraform ran from its official container image on a new AWS configuration.
  • Terraform steps fmt -check, init -backend=false, validate, and providers lock passed on the first try.
🕒 2026-10-06 · new reporting from Hacker News Front Page
  • Berth is a new open-source tool.
  • Berth allows agents to run on remote development boxes.
  • Berth enables agents to continue working when a laptop is closed.
  • Berth is available for Apple silicon and Intel Macs.
  • Berth is signed and notarized.
  • Berth creates a worktree on a chosen box and starts an agent.
  • Berth displays agent commands, reads, and edits as diffs.
  • Berth allows comparing two worktrees in mirrored lanes.
  • Berth offers its own themes and nineteen ported themes.
  • Berth allows customizing chat background and text size.
🕒 2026-10-03 · new reporting from Hacker News Front Page
  • Offrun is a new workspace to manage multiple coding agents.
  • Offrun uses Git worktrees to isolate agent activities.
  • Offrun allows concurrent tasks like refactoring and bug fixes without file conflicts.
  • Offrun integrates peer review and account management features.
  • Offrun runs CLIs on the user's Mac, signed in as the user.
  • Offrun allows picking a reviewer whose findings land in chat while work is uncommitted.
  • Offrun allows pointing each agent at a different login.
  • Offrun moves chat to a login with room when one hits its limit.
  • Offrun keeps repo-wide context for project memory.
🕒 2026-09-26 · new reporting from Hacker News Front Page
  • drawgent integrates AI coding agents with Excalidraw.
  • drawgent allows generating and modifying diagrams using natural language commands.
  • drawgent connects to Claude Code, Codex, or opencode installations.
  • drawgent uses a screenshot and scene to allow agents to look at the canvas.
  • drawgent requires Node.js >= 18 for Claude and Codex bridges.
  • drawgent setup command configures agents.
  • drawgent up command starts a new agent session and canvas.
  • drawgent up --attach connects to a running session.
  • drawgent writes configuration to ~/.config/drawgent/config.toml.
  • opencode speaks ACP itself.
  • Claude Code and Codex use official ACP adapters.
🕒 2026-09-21 · new reporting from The New Stack
  • Grok Build added memory on September 16.
  • Grok's memory uses Markdown files in workspace and global scopes.
  • Claude Code's auto memory keeps a MEMORY.md index plus one file per note, per repository.
  • Claude Code's auto memory is on by default.
  • Claude Code's Projects beta was announced September 17.
  • Claude Code's Projects beta is for select Pro and Max subscribers with no existing projects.
🕒 2026-09-20 · new reporting from Hacker News Front Page
  • A 'Chief of Staff' pattern separates orchestration from execution for long-horizon AI coding agents.
  • The 'Chief of Staff' pattern uses a coordinating session to manage tasks and verify outcomes.
  • The 'Chief of Staff' pattern stores state externally to overcome context limitations and unreliable self-reports.
🕒 2026-09-18 · new reporting from Hacker News Front Page
  • Claude Code reads AGENTS.md if CLAUDE.md is absent.
  • Claude Code added CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1 for egress through a forward proxy.
  • Claude Code added an optional headers map on gateway upstreams for static headers to a proxy.
  • Claude Code fixed sessions hanging after internal errors.
  • Claude Code fixed conversations failing with "text content blocks must be non-empty".
  • Claude Code fixed unexpected logouts when an older build runs on the same machine.
🕒 2026-09-18 · new reporting from Hacker News Front Page, The Verge
  • Skillsync transfers AI chat sessions between different AI coding agents.
  • Skillsync converts proprietary session formats to prevent vendor lock-in.
  • Skillsync collects and makes all user sessions searchable.
  • Skillsync identifies loaded skills used by agents in sessions.
  • Skillsync allows creating shared workspaces for team session syncing.
  • Skillsync's core is an open-source Rust engine called txcript.
  • Claude Code's Projects feature allows managing multiple agents in a single cloud environment.
  • Claude Code Projects enable coordinated AI agent teams with shared memory and goals.
  • Claude Code Projects have 'threads' for parallel tasks and a 'coordinator' for overall progress.
  • Each thread in Claude Code Projects works on its own branch and repo copy.
  • Claude Code Projects resolve overlapping code work as merge conflicts.
  • Claude Code Projects will support local threads in the future.
🕒 2026-09-13 · new reporting from Hacker News Front Page
  • AgentsDock is an IDE for agentic AI research.
  • AgentsDock supports macOS 14+.
  • AgentsDock allows researchers to manage multiple AI agents.
  • AgentsDock allows connecting to various servers.
  • AgentsDock allows accessing development tools from any device.
  • AgentsDock aims to centralize AI research workflows.
  • AgentsDock server is installed via git clone from ZhengyiLuo/AgentsServer.git.
  • AgentsDock allows side-by-side chat with different agents.
  • AgentsDock allows viewing and editing code on the remote.
  • AgentsDock provides full terminal access from anywhere.
🕒 2026-09-09 · new reporting from Hacker News Front Page
  • OtoDock released a self-hosted platform for collaborative agents.
  • OtoDock allows agents to perform tasks, interact with tools, and work autonomously.
  • OtoDock is multi-tenant by design.
  • OtoDock offers four modes for sharing agents and their work.
  • OtoDock provides a self-hosted dashboard and phone line interaction for agents.
  • OtoDock agents run on user's Anthropic and OpenAI subscriptions, and local models.
  • OtoDock agents have a name and a job, like Personal Assistant or System Admin.
  • OtoDock agents are made of six editable parts.
🕒 2026-09-09 · new reporting from Hacker News Front Page
  • Geiger is a new open-source command-line tool.
  • Geiger inventories AI agents, harnesses, MCP servers, plugins, and extensions.
  • Geiger identifies what each AI component can access and execute.
  • Geiger is installed via `npx geiger-scan`.
  • Geiger does not require installation, an account, or telemetry.
  • Geiger reads configs and directories, and writes nothing by default.
  • Geiger can output to a JSON file.
  • An open-source agent harness gained 200,000+ GitHub stars in three weeks in August 2026.
  • The harness's plugin ecosystem passed 13,000 repositories in three weeks.
  • Geiger found 9 findings across 3 ecosystems on a machine.
  • Geiger reported 7 components could execute code and 1 credential was found.
🕒 2026-09-08 · new reporting from Hacker News Front Page
  • Claude Code runs on Firecracker microVMs.
  • Claude Code's VM uses a custom-built guest kernel (6.18.5-fc-v20).
  • Claude Code's VM init is a Rust binary named process_api.
  • process_api listens on vsock port 2024 to allow host control.
🕒 2026-09-03 · new reporting from Hugging Face Blog
  • Funes is a new tool providing a durable, local memory layer for coding agents.
  • Funes indexes agent session traces to allow agents to recall past decisions and rationales.
  • Funes supports Claude Code, Codex, pi, and Hermes agents.
  • Funes can travel to a Hugging Face dataset owned by the user, private by default.
  • Funes is a single binary with no ML runtime dependency for its default inference backend.
🕒 2026-08-21 · new reporting from Hacker News Front Page
  • Proliferate is an open-source, self-hostable platform.
  • Proliferate runs Claude Code, Codex, OpenCode, Cursor, and Grok.
  • Proliferate provides isolated git worktrees for each task.
  • Proliferate supports subagents that delegate work to child agents.
  • Proliferate offers integrations for MCPs, skills, Computer Use, and Browser Use.
  • Proliferate includes workflows for recurring and event-driven agent runs.
  • Proliferate runs agents through their native harnesses.
  • Proliferate can be deployed with Docker, AWS, GCP, Azure, Kubernetes, and air-gapped operation.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

How outlets covered it

IDrive launched AI Agents Backup, a new feature in its desktop application that backs up local data from coding agents like Claude Code, Cursor, Codex, and Cline. This addresses the issue of valuable session transcripts, instruction files, and configurations often stored in hidden local folders and prone to loss, providing a way to preserve developer work product.

Vercel launched Board, an experimental shared message board designed for AI agents to coordinate tasks, share findings, and track progress across different tools. Vercel developed Board for its internal engineering operations, using it to coordinate security agents and rebuild v0's software factory, which now merges over 50% of PRs.

A Windows 11 system DLL, MissionControlAggregator.dll, records usage of developer tools like Python and GCC, along with AI coding agents such as Copilot and Claude. This data is saved locally and reported to Windows Diagnostics, then uploaded to Microsoft if users have not opted out of optional diagnostic data.

Agweb is a new macOS browser designed for running multiple coding agents simultaneously, utilizing WebKit to prevent focus stealing and reduce resource consumption. It allows agents to be driven from the shell while providing a live dashboard view of all agent activities, addressing issues like headless browser blocking and high CPU usage from other automation tools.

AgentRun is an open-source platform that adds self-improvement, durable execution, and a control plane to existing AI agents. It allows agents to learn from past runs, resume execution after interruptions, and provides a unified console for management, aiming to reduce token usage and improve fault tolerance.

Atuin 18.24 introduces Atuin Threads, a feature that allows users to sync, search, and resume AI agent sessions across multiple machines. This update addresses the issue of scattered agent session context by providing a unified history, enabling workflows to transfer between different agents and devices.

Anthropic's AI coding assistant, Claude Code, saves commands containing passwords and API keys in a project settings file when users select "Yes, and don't ask again." This behavior has led to 1,122 public copies of this file on GitHub containing at least one secret, making them accessible to anyone. The issue persists because the file, .claude/settings.local.json, is not consistently excluded from Git repositories, and older versions remain in git history even if the file is removed.

Stroq is a new command-line interface (CLI) tool that audits AI coding agent sessions, such as those from Claude Code and Codex, to identify data leakage and security risks. It analyzes agent activity, including commands executed and data sent, to determine if sensitive information like API keys left the machine or if commands repeated agent-read text.

Inkitt has developed a system that automates small, low-risk software changes from task intake to a draft pull request, shipping code daily. The system uses an event-driven orchestrator that coordinates short-lived agent containers and can resume execution from a recorded state.

A developer reported that the Postman desktop application automatically installed plugins into Claude Code and OpenAI Codex, and attempted to approve its own startup hook, without explicit user consent. This incident raises concerns about third-party applications modifying AI coding agent configurations without clear opt-in mechanisms.

The Postman desktop application has been observed to automatically install AI plugins, including Codex and Claude Code, without explicit user consent. This installation process also attempts to approve its own startup hooks, which are designed to route API engineering tasks through Postman skills. The behavior raises concerns about consent and trust boundaries for an API client.

Claude Code sends each background agent's full transcript to its model every 30 seconds to generate a 3-5 word status update, a process that consumes significant rate-limit headroom. This behavior, active since August 2026, can be disabled using a provided JavaScript preload script.

Corgi is a new tool that functions as both an agent and terminal multiplexer, designed to help users manage multiple agents and their activities. It tracks agent status, organizes workspaces, and provides features like activity tracking and resource usage monitoring.

A new Claude Code mod, agent-usage-guard, prevents rapid consumption of the five-hour usage window by monitoring live usage and controlling subagent fan-out, context churn, and retry loops. It acts before API calls are made, prompting the user for approval or refusing actions to conserve usage.

This guide explains how to use Tmux for managing AI coding agents, particularly with Claude Code. Tmux provides persistent, scriptable sessions for long-running processes, which is beneficial for AI agents.

Sphinx is a new open-source Git hook that integrates with AI coding agents to quiz developers on the functionality of AI-generated code before it is pushed. This tool aims to ensure developers understand the code produced by AI, preventing commits of unreviewed or misunderstood changes.

The Claude Code team has shifted its code review process to rely heavily on automated checks, with Claude AI performing initial reviews. This change allowed the team to more than double merged pull requests while reducing the rate of reverts by over 40% between June and September.

Chainguard released two beta products, Workspaces and Checks, designed to secure development environments for human developers and AI coding agents. These tools address security risks and inefficiencies associated with running AI agents and code changes directly on developer machines or through traditional CI systems.

A test explored how Claude Code can use dependency records from a build platform to understand an application's architecture without prior conversation history or local code. This approach aims to provide context for AI agents beyond traditional memory functions, allowing them to work on existing projects more effectively.

Agent.reviews has launched a new platform where AI agents read and write reviews on various development tools. The platform provides setup steps and documentation for agents to interact with and evaluate tools like Claude Code, Google Cloud Run, Terraform, and GitHub.

Berth is a new open-source tool that allows coding agents like Claude Code and Codex to run on remote development boxes, independent of a local laptop's status. This enables agents to continue working even when the user's laptop is closed, with progress accessible upon reopening.

Offrun launched a new workspace designed to manage multiple coding agents within a single project, utilizing Git worktrees to isolate agent activities. This system allows developers to run concurrent tasks like refactoring and bug fixes without file conflicts, integrating peer review and account management features.

drawgent integrates AI coding agents like Claude Code, Codex, or opencode with Excalidraw, allowing users to generate and modify diagrams using natural language commands. This tool enables developers to create visual representations directly from text prompts or by interacting with existing drawing elements.

Claude Code's AGENTS.md feature, intended to provide project instructions, failed to load when telemetry was disabled due to a remote feature flag dependency. This issue, affecting versions up to 2.1.277, meant the feature was unavailable unless specific environment variables were unset or overridden via session settings. The problem has been addressed in version 2.1.280, where the feature is now enabled by default.

A test compared the memory capabilities of xAI's Grok Build and Anthropic's Claude Code by evaluating their ability to retain information across sessions. The analysis focused on whether these AI coding agents could recall previously provided facts when performing new tasks without explicit reminders.

A new organizational pattern, dubbed 'Chief of Staff', addresses the limitations of long-horizon AI coding agents by separating orchestration from execution. This approach uses a coordinating session to manage tasks and verify outcomes, storing state externally to overcome context limitations and unreliable self-reports in AI coding workflows.

Claude Code now supports reading AGENTS.md files when CLAUDE.md is absent, and introduces new gateway proxy configurations. The update also addresses several bugs, including issues with session hanging, conversation failures, and update checks.

Claude Code relaunched its Projects feature, enabling users to manage multiple AI agents with shared memory, goals, and files within a single cloud environment. This update allows for coordinated AI agent teams, with individual threads handling tasks and a coordinator managing overall progress, aiming to improve efficiency for complex assignments.

Skillsync has launched a new tool that allows users to transfer AI chat sessions, including messages, reasoning, and tool calls, between different AI coding agents. This aims to address the issue of vendor lock-in by converting proprietary session formats, enabling users to switch agents without losing conversational history and accumulated knowledge.

AgentsDock, an IDE for agentic AI research, has been released, offering a unified workspace for AI agents like Claude Code, Codex, and Cursor. It allows researchers to manage multiple AI agents, connect to various servers, and access development tools from any device, aiming to centralize AI research workflows.

OtoDock released a self-hosted platform for collaborative agents, allowing companies to deploy AI agents powered by Anthropic's Claude Code and OpenAI's Codex. The platform enables agents to perform tasks, interact with tools, and work autonomously within a company's operations.

A new open-source command-line tool named Geiger has been released, which inventories all AI agents, harnesses, MCP servers, plugins, and extensions on a machine. The tool identifies what each AI component can access and execute, addressing the growing security concerns around the proliferation of AI agents with broad permissions.

An analysis details the virtual machine (VM) infrastructure enabling mobile AI agents such as Claude Code, highlighting how these agents run on cloud-provided VMs rather than locally on user devices. This approach offloads computational demands from mobile phones and allows AI companies to manage the execution environment.

Funes is a new tool that provides a durable, local memory layer for coding agents like Claude Code, Codex, pi, and Hermes. It indexes agent session traces to allow agents to recall past decisions and rationales, addressing the problem of agents forgetting previous work.

Proliferate is an open-source, self-hostable platform that allows users to run various coding agents like Claude Code, Codex, and Grok in parallel within a single workspace. This tool provides isolated environments for each task, enabling concurrent execution and management of different AI coding assistants.

Wallfacer is a new terminal session manager designed to organize and manage coding sessions from various AI agents, including Claude Code, Cursor CLI, Kiro CLI, and Codex. It addresses the challenge of scattered and unorganized session files by providing a unified interface for searching, grouping, and resuming past interactions, improving developer workflow with AI coding tools.

Agent-Manager is a new terminal user interface (TUI) that allows developers to run and manage multiple AI coding agents like Claude Code, Codex, OpenCode, and Grok within individual tmux sessions. This tool centralizes the monitoring and interaction with these agents, enabling users to send prompts and review changes from a single interface.