Skip to content
Projects

xenodium/agent-shell

View on GitHub

xenodium/agent-shell has 2,038 GitHub stars and 265 forks. GitHot measured 0 stars in 24 hours and +17 in 7 days. Last checked 2026-11-07 20:53 UTC. Last push 2026-11-03.

How these counts are measured

Brings your favorite AI agents to Emacs, keeping you at home: familiar buffers and bindings, no terminal emulation. Powered by ACP.

2,038 stars265 forks

Measured star growth

in 24h
0
in 7 days
+17
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

README.md

Original

#+TITLE: Emacs Agent Shell #+AUTHOR: Álvaro Ramírez

[[https://melpa.org/#/agent-shell][file:https://melpa.org/packages/agent-shell-badge.svg]]

👉 [[https://github.com/sponsors/xenodium][Support this work via GitHub Sponsors]] by [[https://github.com/xenodium][@xenodium]] (check out my [[https://xenodium.com][blog]])

[[file:agent-shell.png]]

  • This project needs your funding

As you pay for those useful LLM tokens, consider [[https://github.com/sponsors/xenodium][sponsoring]] development and maintenance of this project. With your help, I can make this effort more [[https://github.com/sponsors/xenodium][sustainable]].

Thank you!

[[https://xenodium.com/][Alvaro]]

  • agent-shell

=agent-shell= brings your favorite AI agents to [[https://www.gnu.org/software/emacs/][Emacs]], keeping you at home: familiar buffers and bindings, no terminal emulation. All powered by ACP, the [[https://agentclientprotocol.com][Agent Client Protocol]].

Works with 24 agents today, and any other that speaks ACP:

[[https://www.youtube.com/watch?v=R2Ucr3amgGg][agent-shell - YouTube]]

[[https://www.youtube.com/watch?v=R2Ucr3amgGg][file:yt.png]]

[[https://www.youtube.com/watch?v=ymMlftdGx4I][agent-shell + Claude Skills - YouTube]]

[[https://www.youtube.com/watch?v=ymMlftdGx4I][https://img.youtube.com/vi/ymMlftdGx4I/0.jpg]]

[[https://www.youtube.com/watch?v=HJQ86HuSIJI][agent-shell + Claude Skills + Charts - YouTube]]

[[https://www.youtube.com/watch?v=HJQ86HuSIJI][https://img.youtube.com/vi/HJQ86HuSIJI/0.jpg]]

[[https://www.youtube.com/watch?v=7fdHeUoRRgg][Bending Emacs Episode 14: Prototyping iOS apps with agent-shell, artist-mode, and Claude Skills - YouTube]]

[[https://www.youtube.com/watch?v=7fdHeUoRRgg][https://img.youtube.com/vi/7fdHeUoRRgg/0.jpg]]

  • News

Do take a look at all these blog posts as they showcase most =agent-shell= features.

  • Related projects

=agent-shell= relies on [[https://github.com/xenodium/acp.el][acp.el]] to communicate with agents via ACP ([[https://agentclientprotocol.com/][Agent Client Protocol]]).

We now have a handful of additional packages to extend the =agent-shell= experience:

  • Articles
  • Icons

Thanks to [[https://github.com/lobehub/lobe-icons][Lobe Icons]] for the lovely icons.

  • Setup

** External dependencies

*** Claude Agent SDK

For Anthropic's [[https://code.claude.com/docs/en/overview][Claude Agent]] (formerly known as the Claude Code), follow [[https://github.com/agentclientprotocol/claude-agent-acp][claude-agent-acp instructions]], typically something like:

#+begin_src bash npm install -g @agentclientprotocol/claude-agent-acp #+end_src

Note: The =-g= flag is required to install the binary globally so it's available in your PATH. After installation, verify it's available by running =which claude-agent-acp= in your terminal.

Optionally: You may also need [[https://code.claude.com/docs/en/overview][Claude Agent]] itself if you want to use a Claude subscription (run the CLI outside of Emacs at least once to log in to your subscription and then use =agent-shell= from Emacs). Follow Claude Agent's [[https://code.claude.com/docs/en/overview#get-started][get started]] for installation.

*** Codex

For OpenAI's Codex, install [[https://github.com/agentclientprotocol/codex-acp][@agentclientprotocol/codex-acp]] (=npm install -g @agentclientprotocol/codex-acp=) and ensure the =codex-acp= executable is in PATH.

*** Gemini CLI

For Google's [[https://github.com/google-gemini/gemini-cli][Gemini CLI]], be sure to get a recent release supporting the =--experimental-acp= flag.

*** Goose

For Goose CLI, install [[https://goose-docs.ai/docs/getting-started/installation][goose]] and ensure the =goose= executable is in PATH.

*** Grok Build (xAI)

For [[https://docs.x.ai/docs/overview][Grok Build]], install the Grok CLI so =grok= is on =PATH= (the installer typically puts it at =/.grok/bin=). Authenticate once via the CLI login flow (credentials live in =/.grok/auth.json=).

Grok Build speaks ACP over stdio:

#+begin_src bash grok agent stdio #+end_src

Optional flags (set via =agent-shell-xai-acp-command=):

#+begin_src bash grok agent -m grok-build stdio grok agent --always-approve stdio #+end_src

See also the [[https://zed.dev/acp/agent/grok-build][Zed ACP registry listing]] for Grok Build.

*** Cursor

For Cursor agent, install the official [[https://cursor.com/docs/cli][Cursor CLI]] and ensure the =agent= executable is in PATH. =agent-shell= uses its official ACP server (=agent acp=).

A recent Cursor CLI is required (the =acp= subcommand was added in 2026). Older CLIs won't expose =agent acp=.

*** Kimi Code CLI

For Kimi Code CLI, install with:

#+begin_src bash curl -L code.kimi.com/install.sh | bash #+end_src

See https://www.kimi.com/code for details.

*** CodeBuddy

For CodeBuddy agent, install the CodeBuddy CLI and ensure ACP mode is available. For example:

#+begin_src bash codebuddy --acp #+end_src

See https://www.codebuddy.ai/docs/zh/ide/Getting-Started/Installation for details.

*** DeepSeek Harness

For DeepSeek Harness, install with:

#+begin_src bash npm install -g @deepseek-ai/dsh #+end_src

The ACP server reads your API key from =DEEPSEEK_API_KEY=. If Emacs does not inherit it from your shell, set it via =agent-shell-deepseek-authentication=:

#+begin_src emacs-lisp (setq agent-shell-deepseek-authentication (agent-shell-deepseek-make-authentication :api-key "your-key")) #+end_src

agent-shell launches =dsh --profile acp=, which keeps its own settings separate from other dsh profiles. If models configured elsewhere (for example, via the dsh web UI) are missing, copy them over:

#+begin_src bash cp -i ~/.dsh/profiles/web/cordis.patch.yml ~/.dsh/profiles/acp/cordis.patch.yml #+end_src

See https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/acp/acp/README.md for details.

*** Junie

For Junie, install the [[https://junie.jetbrains.com/docs/junie-cli.html][Junie CLI]] and ensure ACP mode is available:

#+begin_src bash junie --acp true #+end_src

See https://junie.jetbrains.com/docs/junie-cli-acp.html for details.

*** Kiro CLI

For Kiro CLI, install with:

#+begin_src bash curl -fsSL https://cli.kiro.dev/install | bash #+end_src

See https://kiro.dev/docs/cli/acp/ for details.

*** Qoder CLI

Install Qoder CLI following the [[https://docs.qoder.com/cli/installation][Qoder installation instructions]], then log in once:

#+begin_src bash qoder login #+end_src

Qoder CLI includes an ACP server. Start it from =agent-shell= with =M-x agent-shell-qoder-start-agent=; the backend launches:

#+begin_src bash qoder --acp #+end_src

Qoder uses the CLI's existing login state, so a Qoder subscription can be used without configuring a separate model API key. For non-interactive authentication, set =QODER_PERSONAL_ACCESS_TOKEN=...= in =agent-shell-qoder-environment=. See the [[https://docs.qoder.com/cli/acp][Qoder ACP documentation]].

*** Qwen Code

For Qwen Code, install with:

#+begin_src bash npm install -g @qwen-code/qwen-code@latest #+end_src

See https://github.com/QwenLM/qwen-code for details.

*** Auggie

For Auggie CLI, install with:

#+begin_src bash npm install -g @augmentcode/auggie #+end_src

See https://docs.augmentcode.com/cli/overview for details. *** Mistral Vibe

For Mistral Vibe, install with:

#+begin_src bash uv tool install mistral-vibe #+end_src

See https://github.com/mistralai/mistral-vibe for details.

*** Factory Droid

For Factory Droid, install Droid:

#+begin_src bash curl -fsSL https://app.factory.ai/cli | sh #+end_src

See https://factory.ai for details.

*** Pi

For Pi coding agent, install the =pi-acp= adapter:

#+begin_src bash npm install -g pi-acp #+end_src

See https://github.com/svkozak/pi-acp for details. *** Oh My Pi (=omp=)

For the =omp= agent, install with:

#+begin_src bash bun install -g @oh-my-pi/pi-coding-agent #+end_src

See https://github.com/can1357/oh-my-pi for details.

*** Antigravity

For Antigravity, download the =agy_acp_server= binary for your platform from the [[https://github.com/agentclientprotocol/registry/tree/main/antigravity-acp][ACP registry entry]] and make sure it's on your =PATH= (or set =agent-shell-antigravity-acp-command= to an absolute path).

Alternatively, you can bootstrap the antigravity tool with the following function:

#+begin_src emacs-lisp (defun agent-shell-antigravity-bootstrap (&optional force) "Download and install the latest Antigravity ACP server binary." (interactive "P") (let* ((arch (pcase (car (split-string system-configuration "-")) ((or "x86_64" "amd64") "x86_64") ((or "aarch64" "arm64") "aarch64") (a a))) (os (pcase system-type ('darwin "darwin") ('windows-nt "windows") (_ "linux"))) (platform (format "%s-%s" os arch)) (install-dir (agent-shell-cache-dir "antigravity" platform)) (bin-name (file-name-nondirectory (car agent-shell-antigravity-acp-command))) (bin-path (expand-file-name bin-name install-dir))) (unless (and (not force) (file-executable-p bin-path)) (message "Antigravity: bootstrapping server for %s..." platform) (make-directory install-dir t) (let* ((reg-url "https://raw.githubusercontent.com/agentclientprotocol/registry/main/antigravity-acp/agent.json") (reg (with-temp-buffer (url-insert-file-contents reg-url) (json-parse-buffer :object-type 'alist))) (archive-url (map-nested-elt reg `(distribution binary ,(intern platform) archive))) (zip (expand-file-name "server.zip" install-dir))) (url-copy-file archive-url zip t) (call-process "unzip" nil nil nil "-o" "-q" zip "-d" install-dir) (delete-file zip) (set-file-modes bin-path #o755))) (setq agent-shell-antigravity-acp-command (cons bin-path (cdr agent-shell-antigravity-acp-command))) (message "Antigravity: using ACP server at %s" bin-path) bin-path)) #+end_src

See https://antigravity.google/docs/ide/extensions for details.

*** OpenClaw

For OpenClaw, install with:

#+begin_src bash curl -fsSL https://openclaw.ai/install.sh | bash #+end_src

The =openclaw acp= bridge requires a running gateway. Start one with =openclaw gateway start= (verify with =openclaw gateway status=).

See https://docs.openclaw.ai/start/getting-started for details.

*** OpenCode + local ollama models

For opencode coding agent, install with:

#+begin_src bash curl -fsSL https://opencode.ai/install | bash #+end_src

See https://opencode.ai/ for details.

For local ollama models, sudo permission is needed to initialize ollama.service. The following will install ollama and a local Qwen2.5:7b model.

#+begin_src bash curl -fsSL https://ollama.com/install.sh | sh ollama pull qwen2.5:7b #+end_src

If you are behind a network proxy, ollama pull will timeout. You then need to add proxy settings inside /etc/systemd/system/ollama.service

See https://ollama.com/ for details

** Installation

=agent-shell= is powered by built-in =comint-shell=, via [[https://github.com/xenodium/shell-maker][shell-maker]], available on [[https://melpa.org/#/shell-maker][MELPA]].

Both [[https://melpa.org/#/agent-shell][agent-shell]] and its dependency [[https://melpa.org/#/acp][acp.el]] are now available on MELPA.

You can install via:

#+begin_src emacs-lisp (use-package agent-shell :ensure t :ensure-system-package ;; Add agent installation configs here ((claude . "brew install claude-code") (claude-agent-acp . "npm install -g @agentclientprotocol/claude-agent-acp"))) #+end_src

This will automatically install the required dependencies ([[https://melpa.org/#/acp][acp.el]] and [[https://melpa.org/#/shell-maker][shell-maker]]).

*** Optional: image utilities

Pasting images from the clipboard (=agent-shell-send-clipboard-image= or =yank= or =C-y=) and capturing screenshots (=agent-shell-send-screenshot=) rely on external tools. Install the ones for your platform:

| Platform | Clipboard paste | Screenshot capture | |-----------------+------------------------------------------+------------------------------| | macOS | =brew install pngpaste= | =screencapture= (preinstalled) | | Linux (Wayland) | =wl-paste= (e.g. =apt install wl-clipboard=) | =import= (from ImageMagick) | | Linux (X11) | =apt install xclip= | =import= (from ImageMagick) | | Windows | =powershell= | (not bundled) |

Handlers are detected at runtime via =executable-find=. See =agent-shell-clipboard-image-handlers= and =agent-shell-screenshot-command= to customize.

*** Doom Emacs

If you are using Doom Emacs and would like to use the =package!= macro:

#+begin_src emacs-lisp (package! shell-maker) (package! acp) (package! agent-shell) #+end_src

Run =doom sync= and restart.

Include =require= before configuration:

#+begin_src emacs-lisp (require 'acp) (require 'agent-shell) ;; rest of config... #+end_src

** Configuration

Configure authentication for the agent providers you want to use.

*** Environment variables

Pass environment variables to the spawned agent process by customizing the agent-shell-*-environment variable with agent-shell-make-environment-variables. The helper accepts key/value pairs and exports them when the agent starts.

#+begin_src emacs-lisp (setq agent-shell-anthropic-claude-environment (agent-shell-make-environment-variables "ANTHROPIC_API_KEY" (auth-source-pass-get 'secret "anthropic-api-key") "HTTPS_PROXY" "http://proxy.example.com:8080")) #+end_src

**** Inheriting environment variables

By default, the agent process starts with a minimal environment. To inherit environment variables from the parent Emacs process, use the :inherit-env t parameter in agent-shell-make-environment-variables:

#+begin_src emacs-lisp (setenv "ANTHROPIC_API_KEY" (auth-source-pass-get 'secret "anthropic-api-key"))

(setq agent-shell-anthropic-claude-environment (agent-shell-make-environment-variables :inherit-env t)) #+end_src

This ensures that environment variables like PATH, HOME, and others from your Emacs session are available to the agent process, while still allowing you to override or add specific variables.

**** Loading environment variables from files

You can load environment variables from .env files using the :load-env parameter. This supports both single and multiple files:

#+begin_src emacs-lisp ;; Load from a single .env file (setq agent-shell-anthropic-claude-environment (agent-shell-make-environment-variables :load-env "~/.env" "CUSTOM_VAR" "custom_value"))

;; Load from multiple .env files (setq agent-shell-anthropic-claude-environment (agent-shell-make-environment-variables :load-env '("~/.env" ".env.local") :inherit-env t)) #+end_src

The .env files should contain variables in the format KEY=value, with one variable per line. Comments (lines starting with #) and empty lines are ignored.

*** Anthropic Claude

For login-based authentication (default):

#+begin_src emacs-lisp (setq agent-shell-anthropic-authentication (agent-shell-anthropic-make-authentication :login t)) #+end_src

For API key authentication:

#+begin_src emacs-lisp ;; With string (setq agent-shell-anthropic-authentication (agent-shell-anthropic-make-authentication :api-key "your-anthropic-api-key-here"))

;; With function (setq agent-shell-anthropic-authentication (agent-shell-anthropic-make-authentication :api-key (lambda () (auth-source-pass-get 'secret "anthropic-api-key")))) #+end_src

For OAuth token authentication (the =CLAUDE_CODE_OAUTH_TOKEN= we get from =claude setup-token=):

#+begin_src emacs-lisp ;; With string (setq agent-shell-anthropic-authentication (agent-shell-anthropic-make-authentication :oauth "your-oauth-token-here"))

;; With function (setq agent-shell-anthropic-authentication (agent-shell-anthropic-make-authentication :oauth (lambda () (auth-source-pass-get "secret" "anthropic-oauth-token")))) #+end_src

For alternative Anthropic-compatible API endpoints, configure via environment variables:

#+begin_src emacs-lisp (setq agent-shell-anthropic-claude-environment (agent-shell-make-environment-variables "ANTHROPIC_BASE_URL" "https://api.moonshot.cn/anthropic" "ANTHROPIC_MODEL" "kimi-k2-turbo-preview" "ANTHROPIC_SMALL_FAST_MODEL" "kimi-k2-turbo-preview")) #+end_src

*** Antigravity

For login-based authentication (default, Google account):

#+begin_src emacs-lisp (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :login t)) #+end_src

For API key authentication:

#+begin_src emacs-lisp ;; With string (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :api-key "your-gemini-api-key-here"))

;; With function (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :api-key (lambda () (auth-source-pass-get 'secret "gemini-api-key")))) #+end_src

For Gemini Enterprise authentication:

#+begin_src emacs-lisp (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :business t)) #+end_src

For Gemini Enterprise Agent Platform authentication:

#+begin_src emacs-lisp (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :agent-platform t)) #+end_src

For no authentication (when managed externally):

#+begin_src emacs-lisp (setq agent-shell-antigravity-authentication (agent-shell-antigravity-make-authentication :none t)) #+end_src

*** Google Gemini

For login-based authentication (default):

#+begin_src emacs-lisp (setq agent-shell-google-authentication (agent-shell-google-make-authentication :login t)) #+end_src

For API key authentication:

#+begin_src emacs-lisp ;; With string (setq agent-shell-google-authentication (agent-shell-google-make-authentication :api-key "your-google-api-key-here"))

;; With function (setq agent-shell-google-authentication (agent-shell-google-make-authentication :api-key (lambda () (auth-source-pass-get 'secret "google-api-key")))) #+end_src

For Vertex AI authentication:

#+begin_src emacs-lisp (setq agent-shell-google-authentication (agent-shell-google-make-authentication :vertex-ai t)) #+end_src

*** OpenAI Codex

For login-based authentication (default):

#+begin_src emacs-lisp (setq agent-shell-openai-authentication (agent-shell-openai-make-authentication :login t)) #+end_src

For API key authentication:

#+begin_src emacs-lisp ;; With string (setq agent-shell-openai-authentication (agent-shell-openai-make-authentication :api-key "your-openai-api-key-here"))

;; With function (setq agent-shell-openai-authentication (agent-shell-openai-make-authentication :api-key (lambda () (auth-source-pass-get 'secret "openai-api-key")))) #+end_src

*** Goose

By default, authentication is handled by Goose's built-in configuration. This needs no =agent-shell= configuration.

Optionally, pass an OpenAI API key from =agent-shell= instead:

#+begin_src emacs-lisp ;; With string (setq agent-shell-goose-authentication (agent-shell-make-goose-authentication :openai-api-key "your-openai-api-key-here"))

;; With function (setq agent-shell-goose-authentication (agent-shell-make-goose-authentication :openai-api-key (lambda () (auth-source-pass-get 'secret "openai-api-key")))) #+end_src

*** Qwen Code

For OAuth login-based authentication:

#+begin_src emacs-lisp (setq agent-shell-qwen-authentication (agent-shell-qwen-make-authentication :login t)) #+end_src

For an OpenAI-compatible provider (for example OpenRouter), pass the API key and set the base URL and model via =agent-shell-qwen-environment=:

#+begin_src emacs-lisp ;; With string (setq agent-shell-qwen-authentication (agent-shell-qwen-make-authentication :openai-api-key "your-openai-api-key-here"))

;; With function (setq agent-shell-qwen-authentication (agent-shell-qwen-make-authentication :openai-api-key (lambda () (auth-source-pass-get 'secret "openrouter-api-key"))))

(setq agent-shell-qwen-environment (agent-shell-make-environment-variables "OPENAI_BASE_URL" "https://openrouter.ai/api/v1" "OPENAI_MODEL" "x-ai/grok-4.3")) #+end_src

*** Cursor

By default authentication is handled externally: run =agent login= once outside Emacs and agent-shell uses that existing login (it sends no ACP authenticate request). This needs no configuration.

Optionally, have agent-shell drive login-based authentication (=cursor_login=):

#+begin_src emacs-lisp (setq agent-shell-cursor-authentication (agent-shell-cursor-make-authentication :login t)) #+end_src

For API key authentication (passed via the =CURSOR_API_KEY= environment variable):

#+begin_src emacs-lisp ;; With string (setq agent-shell-cursor-authentication (agent-shell-cursor-make-authentication :api-key "your-cursor-api-key-here"))

;; With function (setq agent-shell-cursor-authentication (agent-shell-cursor-make-authentication :api-key (lambda () (auth-source-pass-get 'secret "cursor-api-key")))) #+end_src

For auth token authentication (passed via the =CURSOR_AUTH_TOKEN= environment variable):

#+begin_src emacs-lisp (setq agent-shell-cursor-authentication (agent-shell-cursor-make-authentication :auth-token "your-cursor-auth-token")) #+end_src

For no authentication (already authenticated externally):

#+begin_src emacs-lisp (setq agent-shell-cursor-authentication (agent-shell-cursor-make-authentication :none t)) #+end_src

*** Auggie

For login-based authentication (default):

#+begin_src emacs-lisp (setq agent-shell-auggie-authentication (agent-shell-make-auggie-authentication :login t)) #+end_src

For no authentication (when using alternative authentication methods):

#+begin_src emacs-lisp (setq agent-shell-auggie-authentication (agent-shell-make-auggie-authentication :none t)) #+end_src

*** Mistral Vibe

For API key authentication:

#+begin_src emacs-lisp ;; With string (setq agent-shell-mistral-authentication (agent-shell-mistral-make-authentication :api-key "your-mistral-api-key-here"))

;; With function (reusing the API key configured in vibe) (setq agent-shell-mistral-authentication (agent-shell-mistral-make-authentication :api-key (lambda () (string-trim (shell-command-to-string "source ~/.vibe/.env; echo $MISTRAL_API_KEY"))))) #+end_src

*** OpenCode + local ollama Qwen2.5

If opencode is not found by agent-shell, you can add it to the search path

#+begin_src emacs-lisp (use-package agent-shell :ensure t :config (add-to-list 'exec-path "~/.opencode/bin/")) #+end_src

To register the local Qwen2.5:7b model in opencode, two files needs to be edited:

/.config/opencode/opencode.json #+BEGIN_SRC js { "$schema": "https://opencode.ai/config.json", "permission": { "edit": "ask", "bash": "ask", "webfetch": "allow" }, "provider": { "ollama": { "name": "qwen2.5:7b", "npm": "@ai-sdk/openai-compatible", "options": { "baseURL": "http://localhost:11434/v1" }, "models": { "qwen2.5:7b": { "name": "Qwen2.5 7b" } } } } } #+END_SRC

and a dummy auth file /.local/share/opencode/auth.json

#+begin_src js { "ollama": { "type": "api", "key": "ollama" } } #+end_src

Huge thanks for /gdindi/ for finding and and describing this in https://github.com/xenodium/agent-shell/issues/526

*** Customizing Available Agents

By default, =agent-shell= includes configurations for all supported agents (Claude Agent, Gemini CLI, Codex, Goose, Grok Build, Cursor, CodeBuddy, Qwen Code, and Auggie). You can customize which agents are available through the =agent-shell-agent-configs= variable.

*** Handling provider notifications

Provider implementations can set =:notification-handler= in =agent-shell-make-agent-config= to handle custom ACP notifications. The callback runs after =:notification-adapter= and receives an event alist containing =:state= and the adapted =:acp-notification=. Return non-nil when handled, or nil to continue with core notification handling.

#+begin_src emacs-lisp (agent-shell-make-agent-config :identifier 'example :notification-handler (lambda (event) (when (equal (map-nested-elt event '(:acp-notification method)) "_example/status") (message "%s" (map-nested-elt event '(:acp-notification params text))) t))) #+end_src

Return nil for standard notifications such as =session/update= so core keeps rendering them. See =C-h f agent-shell-make-agent-config= for the configuration options.

** Usage

*** Quick Start

=M-x agent-shell= - Start or reuse any of the known agents.

You can select and start any of the known agent shells (see =agent-shell-agent-configs=) via the =agent-shell= interactive command and enables reusing existing shells when available. With a prefix argument (=C-u M-x agent-shell=), it forces starting a new shell session, thus instantiating multiple agent shells.

*** Viewport interaction

Viewport is an alternative interface for composing prompts and viewing one request/response interaction at a time. It uses a separate buffer named after its associated shell with =" [viewport]"= appended. This buffer switches between =agent-shell-viewport-edit-mode= while composing and =agent-shell-viewport-view-mode= while reading.

Use one of these entry points to create and initialize the viewport buffer:

  • From an agent shell, type =C-c C-o= (=agent-shell-other-buffer=). At the live prompt this opens the viewport for composing; from an earlier interaction it opens the viewport at that interaction. Use =C-c C-o= again to return to the shell (=o= also works from viewport view mode).
  • Run =M-x agent-shell-prompt-compose= to compose in a dedicated buffer. With the default configuration, =C-c C-c= sends or queues the prompt and dismisses the compose window.
  • Set =agent-shell-prefer-viewport-interaction= to make the normal =agent-shell= command and related commands prefer the viewport:

#+begin_src emacs-lisp (setopt agent-shell-prefer-viewport-interaction t) #+end_src

In edit mode, =C-c C-c= sends or queues the prompt and switches to view mode when viewport interaction is preferred. Use =C-u C-c C-c= to send or queue while keeping the compose buffer open, =C-c C-p= to peek at the latest interaction without losing the draft, =C-c C-k= to cancel, and =C-c C-h= for help.

In view mode, use =r= to compose a reply, =R= to quote the response in a reply, =b= / =f= to move between interactions, =p= / =n= to move between items, =?= for help, and =q= to bury the viewport.

#+begin_quote [!NOTE]

Do not invoke =agent-shell-viewport-edit-mode= or =agent-shell-viewport-view-mode= directly in an agent shell buffer. They are major modes installed by the entry points above on the separate viewport buffer, not commands for converting the shell buffer. #+end_quote

*** Typing while the agent is busy

When =agent-shell-persistent-prompt-enabled= is set, the shell keeps a writable prompt at the end of the buffer at all times (including agent busy times).

Submitting into a busy agent queues the prompt by default. =RET= submits via =agent-shell-busy-submit-default-function= (queues), while =M-RET= and =C-u RET= submit via =agent-shell-busy-submit-override-function= (steers). Steering is covered in the next section.

To disable/hide the persistent prompt, unset the related option:

#+begin_src emacs-lisp (setopt agent-shell-persistent-prompt-enabled nil) #+end_src

The viewport's compose buffer is available either way, and submitting from it mid-turn routes the same.

*** Steering a running turn

A prompt sent while the agent is working is queued and submitted once the turn ends. Prompts queued during the same turn are merged into a single prompt, separated by blank lines (set =agent-shell-prompt-queue-merge= to nil to send them one turn each). Cancelling the turn asks whether to continue with the queued prompts. Agents that support steering can take it into the turn already running instead, so a wrong direction can be corrected without waiting for it to finish. Steering is an ACP extension (=_session/steering=) rather than part of the spec, advertised by the agent at initialization: the Claude and Codex packages implement it, other agents keep queueing.

Steering is always explicit. =M-x agent-shell-prompt-queue= queues, as it always has. =M-x agent-shell-prompt-steer= steers the running turn, and errors when the agent cannot steer. With no turn running there is nothing to steer into, so it submits the prompt instead. When the agent declines a steer, the running turn is interrupted and the shell says so, rather than letting the agent carry on in a direction you believe you already corrected.

What the agent does with the steered prompt is up to the agent: Claude interrupts what it is generating and responds to it, while Codex finishes the message in flight first.

Steering is not additive. An agent that interrupts may drop the instruction it was working on, so a steer can replace what the agent was already doing rather than adding to it. Queue instead when the agent should finish first.

*** Specific Agent Commands

Start a specific agent shell session directly:

  • =M-x agent-shell-antigravity-start-agent= - Start an Antigravity agent session
  • =M-x agent-shell-anthropic-start-claude-code= - Start a Claude Agent session
  • =M-x agent-shell-auggie-start-agent= - Start an Auggie agent session
  • =M-x agent-shell-codebuddy-start-agent= - Start a CodeBuddy agent session
  • =M-x agent-shell-openai-start-codex= - Start a Codex agent session
  • =M-x agent-shell-google-start-gemini= - Start a Gemini agent session
  • =M-x agent-shell-goose-start-agent= - Start a Goose agent session
  • =M-x agent-shell-xai-start-grok= - Start a Grok Build (xAI) agent session
  • =M-x agent-shell-cursor-start-agent= - Start a Cursor agent session
  • =M-x agent-shell-deepseek-start-agent= - Start a DeepSeek Harness agent session
  • =M-x agent-shell-junie-start-agent= - Start a Junie agent session
  • =M-x agent-shell-kiro-start-agent= - Start a Kiro CLI agent session
  • =M-x agent-shell-qoder-start-agent= - Start a Qoder CLI agent session
  • =M-x agent-shell-mistral-start-vibe= - Start a Mistral Vibe agent session
  • =M-x agent-shell-qwen-start= - Start a Qwen Code agent session
  • =M-x agent-shell-droid-start-agent= - Start a Factory Droid agent session
  • =M-x agent-shell-pi-start-agent= - Start a Pi coding agent session

*** Setting a default agent for all new shells

You can set a default agent to use for all new shells started via =agent-shell= like so:

#+begin_src emacs-lisp (setq agent-shell-preferred-agent-config (agent-shell-anthropic-make-claude-code-config)) #+end_src

Or by identifier (e.g. Grok Build):

#+begin_src emacs-lisp (setq agent-shell-preferred-agent-config 'grok-build) #+end_src

*** Configuring MCP servers

You can configure MCP servers directly via =agent-shell=. This allows you to avoid having to repeat configurations across every agent that you use.

#+begin_src emacs-lisp (setq agent-shell-mcp-servers '(((name . "notion") (type . "http") (headers . []) (url . "https://mcp.notion.com/mcp")))) #+end_src

*** Monitoring events

You can subscribe to shell events (turn complete, tool calls, config changes, etc.) via =agent-shell-subscribe-to=. See its docstring (=C-h f agent-shell-subscribe-to=) for the full list of events and their data.

For example, to have new OpenCode shells remember the last model and effort level chosen (within your Emacs session):

#+begin_src emacs-lisp (add-hook 'agent-shell-mode-hook (lambda () (agent-shell-subscribe-to :shell-buffer (current-buffer) :event 'config-option-update :on-event (lambda (event) (pcase (map-elt (agent-shell-get-config (current-buffer)) :identifier) ('opencode (setq agent-shell-opencode-default-config-options (seq-map (lambda (option) (cons (map-elt option :id) (map-elt option :current-value))) (map-nested-elt event '(:data :config-options)))))))))) #+end_src

** Running agents in Devcontainers / Docker containers (Experimental)

=agent-shell= provides rudimentary support for running agents and shell commands in containers.

Use =agent-shell-command-prefix= to prefix the command that starts the agent, or a shell command that should be run so it is executed inside the container.

*** Static command list

#+begin_src emacs-lisp (setq agent-shell-command-prefix '("devcontainer" "exec" "--workspace-folder" ".")) #+end_src

*** Function-based configuration

For dynamic per-agent containers, provide a function that takes the current agent-shell buffer and returns the command list:

#+begin_src emacs-lisp (setq agent-shell-command-prefix (lambda (buffer) (let ((config (agent-shell-get-config buffer))) (pcase (map-elt config :identifier) ('claude-code '("docker" "exec" "claude-dev" "--")) ('gemini-cli '("docker" "exec" "gemini-dev" "--")) (_ '("devcontainer" "exec" ".")))))) #+end_src

*** Per-session containers

You can use different containers for different shell sessions, even of the same agent type:

#+begin_src emacs-lisp (setq agent-shell-command-prefix (lambda (buffer) ;; Different container based on project (if (string-match "project-a" (buffer-name buffer)) '("docker" "exec" "project-a-dev" "--") '("docker" "exec" "project-b-dev" "--")))) #+end_src

Note that any =:environment-variables= you may have passed to =acp-make-client= will not apply to the agent process running inside the container. It's expected to inject environment variables by means of your devcontainer configuration / Dockerfile.

Next, set an =agent-shell-path-resolver-function= that resolves container paths in the local working directory, and vice versa. Agent shell provides the =agent-shell-devcontainer-resolve-path= function for use with devcontainers specifically: it reads the =workspaceFolder= specified in =.devcontainer/devcontainer.json=, or uses the default value of =/workspaces/= otherwise.

#+begin_src emacs-lisp (setq agent-shell-path-resolver-function #'agent-shell-devcontainer-resolve-path) #+end_src

Note that this allows the agent to access files on your local file-system. While care has been taken to restrict access to files in the local working directory, it's probably possible for a malicious agent to circumvent this restriction.

Optional: to prevent the agent running inside the container to access your local file-system altogether and to have it read/modify files inside the container directly, in addition to setting the resolver function, disable the "read/write text file" client capabilities:

#+begin_src emacs-lisp (setq agent-shell-text-file-capabilities nil) #+end_src

*** Data storage location

By default, agent-shell stores per-project data (transcripts, screenshots, etc.) under a =.agent-shell/= directory in the project root, and automatically adds that directory to =.gitignore= the first time it is created. This is a one-time operation: removing the entry from =.gitignore= will not cause it to be re-added. If you configure =agent-shell-dot-subdir-function= to store data outside the project =.agent-shell/= directory, agent-shell will not update the project's =.gitignore=.

You can change where this data is stored by setting =agent-shell-dot-subdir-function= to a custom function. The function receives one argument, SUBDIR (e.g. ="screenshots"=), and must return the absolute path to that subdirectory (without creating it).

For example, to store data under =user-emacs-directory= instead of the project tree, flattening the project path into a directory name:

#+begin_src emacs-lisp (defun my/agent-shell-dot-subdir (subdir) (let* ((cwd (string-remove-suffix "/" (agent-shell-cwd))) (sanitized (replace-regexp-in-string "/" "-" (string-remove-prefix "/" cwd)))) (expand-file-name subdir (locate-user-emacs-file (concat "agent-shell/" sanitized)))))

(setopt agent-shell-dot-subdir-function #'my/agent-shell-dot-subdir) #+end_src

This stores data at a path like =~/.emacs.d/agent-shell/home-user-src-myproject/screenshots/=.

*** Screenshots from clipboard

You can send a screenshot from your clipboard to your shell with =agent-shell-send-clipboard-image=. Call with =C-u= to =agent-shell-send-clipboard-image-to= to select from your shells. agent-shell relies on external programs to write an image from clipboard to file as configured in =agent-shell-clipboard-image-handlers=. Preconfigured handlers are

  • =wl-paste= for Wayland desktops,
  • =xclip= for Xorg,
  • =pngpaste= for MacOS.
  • =powershell= for Windows.

Alternatively, =M-x yank-media= pastes a clipboard image with no external programs, where your Emacs build can read images off the clipboard.

*** Dragging files in

Drag a file from your file manager (or =dired=) onto a shell or viewport buffer to attach it to the prompt, the same way a pasted image is. Images get a preview; any other file is attached as an =@path= link.

*** Inhibiting minor modes during file writes

Some minor modes (for example, =aggressive-indent-mode=) can interfere with an agent's edits. Agent Shell can temporarily disable selected per-buffer minor modes while applying edits.

#+begin_src emacs-lisp (setopt agent-shell-write-inhibit-minor-modes '(aggressive-indent-mode)) #+end_src

*** Keeping the system awake during long turns

Long agent turns can outlast the system idle-sleep timeout, suspending the machine (and the agent) before the turn completes. On Emacs 31.1 or later, =agent-shell-inhibit-system-sleep= (enabled by default) keeps the system awake for the duration of each turn and releases the block once the turn finishes, so the system is only kept awake while there is work in progress (the display is still allowed to blank). This relies on the built-in [[https://git.savannah.gnu.org/cgit/emacs.git/tree/lisp/system-sleep.el][=system-sleep=]] library and has no effect on earlier Emacs versions. It does not prevent "hard" sleep such as closing a laptop lid.

To turn it off:

#+begin_src emacs-lisp (setopt agent-shell-inhibit-system-sleep nil) #+end_src

All of the above settings can be applied on a per-project basis using [[https://www.gnu.org/software/emacs/manual/html_node/emacs/Directory-Variables.html][directory-local variables]].

** Key bindings

  • =RET= - Submits prompt.

  • =M-J= - Iserts newline.

  • =C-c C-c= - Interrupt current agent operation

  • =TAB and Shift-TAB= - Navigate interactive elements

    To customize =RET= binding behaviour, you can use something like:

    #+begin_src emacs-lisp :lexical no (use-package agent-shell :bind (:map agent-shell-mode-map ("RET" . newline) ("M-RET" . shell-maker-submit))) #+end_src

    Or if you prefer C-c C-c to send and C-c C-k to interrupt, use the following:

    #+begin_src emacs-lisp :lexical no (use-package agent-shell :bind (:map agent-shell-mode-map ("RET" . newline) ("C-c C-c" . shell-maker-submit) ("C-c C-k" . agent-shell-interrupt))) #+end_src

*** Folding sections

Thoughts, tool calls, user messages and activity groups fold under their header line. With point on a header:

  • =RET= (or click) - Expand or collapse the section.
  • =C-u RET= - Expand or collapse the section's activity group, along with every section inside it.

These keys only apply on headers, so =RET= still submits the prompt everywhere else. To fold with other keys, add them to =agent-shell-ui-fragment-map=.

Use =M-x agent-shell-ui-toggle-all-fragments= to expand or collapse every section in the buffer.

*** Session controls

When the active agent advertises the corresponding session option, you can switch it from the shell buffer:

  • =C-c C-v= - Set model.
  • =C-c C-m= - Set session mode.
  • =C-c C-t= - Set thought level (reasoning effort).
  • =C-= - Cycle session mode.

The current selections are shown in the header right after the agent name; click any segment to open a popup menu. The thought level segment only appears for models that expose it (e.g. Claude Sonnet/Opus with effort levels; Codex's reasoning_effort).

Any other options the agent advertises (e.g. Codex's speed) are counted after the header settings as =(+N)=; click it to change them. They are also appended to the graphical header's settings menu. =C-c C-s= sets any of them from the keyboard.

From the viewport (=agent-shell-viewport-view-mode= / =agent-shell-viewport-edit-mode=), the same controls are bound to:

  • View mode: =v= (model), =s= (mode), =t= (thought level).
  • Edit mode: =C-c C-v=, =C-c C-m=, =C-c C-t=.

*** Evil Evil users may want to rebind RET for inserting a new line in =insert mode= and sending the prompt in =normal mode=.

Also, when viewing diffs (before accepting changes) it may be annoying having to enter =insert mode= to send keys (y/n/p/q/etc). If this is your case, you can make these buffers start in Emacs mode (you can always go to Evil modes if you need to with C-z). #+begin_src emacs-lisp (use-package agent-shell :config ;; Evil state-specific RET behavior: insert mode = newline, normal mode = send (evil-define-key 'insert agent-shell-mode-map (kbd "RET") #'newline) (evil-define-key 'normal agent-shell-mode-map (kbd "RET") #'comint-send-input)

;; Configure *agent-shell-diff* buffers to start in Emacs state
(add-hook 'diff-mode-hook
    (lambda ()
      (when (string-match-p "\\*agent-shell-diff\\*" (buffer-name))
	(evil-emacs-state)))))

#+end_src

  • Customizations

#+BEGIN_SRC emacs-lisp :results table :colnames '("Custom variable" "Description") :exports results (let ((rows)) (mapatoms (lambda (symbol) (when (and (string-match "^agent-shell" (symbol-name symbol)) (custom-variable-p symbol)) (push `(,symbol ,(car (split-string (or (documentation-property symbol 'variable-documentation) (get (indirect-variable symbol) 'variable-documentation) (get symbol 'variable-documentation) "") "\n"))) rows)))) (sort rows (lambda (a b) (string< (symbol-name (car a)) (symbol-name (car b)))))) #+END_SRC

#+RESULTS: | Custom variable | Description | |-----------------------------------------------+---------------------------------------------------------------------------------| | agent-shell-activity-group-expand-by-default | When activity group sections should be expanded. | | agent-shell-agent-configs | The known agent configurations. | | agent-shell-anthropic-authentication | Configuration for Anthropic authentication. | | agent-shell-anthropic-claude-acp-command | Command and parameters for the Anthropic Claude client. | | agent-shell-anthropic-claude-environment | Environment variables for the Anthropic Claude client. | | agent-shell-anthropic-default-model-id | Default Anthropic model ID. | | agent-shell-anthropic-default-session-mode-id | Default Anthropic session mode ID. | | agent-shell-antigravity-acp-command | Command and parameters for the Antigravity ACP server. | | agent-shell-antigravity-authentication | Configuration for Antigravity authentication. | | agent-shell-antigravity-environment | Environment variables for the Antigravity ACP server. | | agent-shell-artist-mode-hook | Hook run after entering or leaving ‘agent-shell-artist-mode’. | | agent-shell-auggie-acp-command | Command and parameters for the Auggie client. | | agent-shell-auggie-authentication | Configuration for Auggie authentication. | | agent-shell-auggie-environment | Environment variables for the Auggie client. | | agent-shell-buffer-name-format | Format to use when generating agent shell buffer names. | | agent-shell-busy-indicator-frames | Frames for the busy indicator animation. | | agent-shell-busy-submit-default-function | Function deciding what submitting while the agent is working does. | | agent-shell-busy-submit-override-function | Function M-x agent-shell-submit-override submits through. | | agent-shell-chat-label-function | Function returning a chat label ("Me" or the agent's name). | | agent-shell-chat-mode-enabled | Whether a new agent shell enables ‘agent-shell-chat-mode’ by default. | | agent-shell-chat-mode-hook | Hook run after entering or leaving ‘agent-shell-chat-mode’. | | agent-shell-cline-acp-command | Command and parameters for the Cline agent client. | | agent-shell-cline-default-model-id | Default Cline model ID. | | agent-shell-cline-default-session-mode-id | Default Cline session mode ID. | | agent-shell-cline-environment | Environment variables for the Cline agent client. | | agent-shell-clipboard-image-handlers | Handlers for saving clipboard images to a file. | | agent-shell-codebuddy-acp-command | Command and parameters for the CodeBuddy agent client. | | agent-shell-codebuddy-environment | Environment variables for the CodeBuddy agent client. | | agent-shell-command-completion-enabled | Non-nil pops agent command completion as / is typed. | | agent-shell-command-prefix | Prefix to apply when executing agent commands and shell commands. | | agent-shell-completion-mode-hook | Hook run after entering or leaving ‘agent-shell-completion-mode’. | | agent-shell-confirm-interrupt | Whether to prompt for confirmation before interrupting. | | agent-shell-container-command-runner | Prefix to apply when executing agent commands and shell commands. | | agent-shell-context-sources | Sources to consider when determining C-c C-w automatic context. | | agent-shell-cursor-acp-command | Command and parameters for the Cursor agent client. | | agent-shell-cursor-authentication | Configuration for Cursor authentication. | | agent-shell-cursor-environment | Environment variables for the Cursor agent client. | | agent-shell-cwd-function | When non-nil, a function called to determine the shell’s CWD. | | agent-shell-deepseek-acp-command | Command and parameters for the DeepSeek Harness ACP server. | | agent-shell-deepseek-authentication | Configuration for DeepSeek Harness authentication. | | agent-shell-deepseek-default-config-options | Default DeepSeek Harness config options, applied at session start. | | agent-shell-deepseek-environment | Environment variables for the DeepSeek Harness ACP server. | | agent-shell-display-action | Display action for agent shell buffers. | | agent-shell-dot-subdir-function | Function used by ‘agent-shell--dot-subdir’ to resolve subdirectory paths. | | agent-shell-droid-acp-command | Command and parameters for the Factory Droid ACP client. | | agent-shell-droid-authentication | Configuration for Factory Droid authentication. | | agent-shell-droid-default-model-id | Default Droid AI model ID. | | agent-shell-droid-default-reasoning-effort | Default reasoning effort. | | agent-shell-droid-default-session-mode-id | Default Droid AI session mode ID. | | agent-shell-droid-environment | Environment variables for the Factory Droid ACP client. | | agent-shell-embed-file-size-limit | Maximum file size in bytes for embedding with ContentBlock::Resource. | | agent-shell-file-completion-enabled | Non-nil pops project file completion as @ is typed. | | agent-shell-file-display-action | Display action for files opened from a link, image, or ‘@file’ mention. | | agent-shell-github-acp-command | Command and parameters for the GitHub Copilot agent client. | | agent-shell-github-default-model-id | Default GitHub Copilot model ID. | | agent-shell-github-default-session-mode-id | Default GitHub Copilot session mode ID. | | agent-shell-github-environment | Environment variables for the GitHub Copilot agent client. | | agent-shell-google-authentication | Configuration for Google authentication. | | agent-shell-google-gemini-acp-command | Command and parameters for the Gemini client. | | agent-shell-google-gemini-environment | Environment variables for the Google Gemini client. | | agent-shell-goose-acp-command | Command and parameters for the Goose client. | | agent-shell-goose-authentication | Configuration for Goose authentication. | | agent-shell-goose-environment | Environment variables for the Goose client. | | agent-shell-header-style | Style for agent shell buffer headers. | | agent-shell-hermes-acp-command | Command and parameters for the Hermes agent client. | | agent-shell-hermes-default-session-mode-id | Default ACP session mode for the Hermes agent. | | agent-shell-hermes-environment | Environment variables for the Hermes agent client. | | agent-shell-highlight-blocks | Whether or not to highlight source blocks. | | agent-shell-inhibit-system-sleep | Non-nil to keep the system awake while an agent is busy. | | agent-shell-junie-acp-command | Command and parameters for the JetBrains Junie ACP client. | | agent-shell-junie-default-model-id | Default Junie model ID. | | agent-shell-junie-default-session-mode-id | Default Junie session mode ID. | | agent-shell-junie-environment | Environment variables for the Junie ACP client. | | agent-shell-kimi-acp-command | Command and parameters for the Kimi Code CLI client. | | agent-shell-kimi-default-model-id | Default Kimi model ID. | | agent-shell-kimi-default-session-mode-id | Default Kimi Code session mode ID. | | agent-shell-kimi-environment | Environment variables for the Kimi Code CLI client. | | agent-shell-kiro-acp-command | Command and parameters for the Kiro CLI client. | | agent-shell-kiro-default-model-id | Default Kiro model ID. | | agent-shell-kiro-default-session-mode-id | Default Kiro session mode ID. | | agent-shell-kiro-environment | Environment variables for the Kiro CLI client. | | agent-shell-list-edit-mode-hook | Hook run after entering or leaving ‘agent-shell-list-edit-mode’. | | agent-shell-markdown-render-function | Function called to render markdown in the current narrowed buffer. | | agent-shell-mcp-servers | List of MCP servers to initialize when creating a new session. | | agent-shell-mistral-acp-command | Command and parameters for the Mistral Vibe client. | | agent-shell-mistral-authentication | Configuration for Mistral AI authentication. | | agent-shell-mistral-default-model-id | Default Mistral AI model ID. | | agent-shell-mistral-default-session-mode-id | Default Mistral AI session mode ID. | | agent-shell-mistral-environment | Environment variables for the Mistral Vibe client. | | agent-shell-mock-agent-acp-command | Command and parameters for the mock ACP agent. | | agent-shell-omp-acp-command | Command and parameters for the Oh My Pi (omp) ACP client. | | agent-shell-omp-environment | Environment variables for the Oh My Pi (omp) client. | | agent-shell-openai-authentication | Configuration for OpenAI authentication. | | agent-shell-openai-codex-acp-command | Command and parameters for the OpenAI Codex client. | | agent-shell-openai-codex-environment | Environment variables for the OpenAI Codex client. | | agent-shell-openai-default-config-options | Default Codex config options, applied at session start. | | agent-shell-openai-default-model-id | Default Codex model ID. | | agent-shell-openai-default-session-mode-id | Default Codex session mode ID. | | agent-shell-openclaw-acp-command | Command and parameters for the OpenClaw ACP bridge. | | agent-shell-openclaw-default-model-id | Default OpenClaw model ID. | | agent-shell-openclaw-environment | Environment variables for the OpenClaw client. | | agent-shell-opencode-acp-command | Command and parameters for the OpenCode client. | | agent-shell-opencode-authentication | Configuration for OpenCode authentication. | | agent-shell-opencode-default-config-options | Default OpenCode config options, applied at session start. | | agent-shell-opencode-default-model-id | Default OpenCode model ID. | | agent-shell-opencode-default-session-mode-id | Default OpenCode session mode ID. | | agent-shell-opencode-environment | Environment variables for the OpenCode client. | | agent-shell-outgoing-request-decorator | Function to decorate outgoing ACP requests before they are sent. | | agent-shell-path-resolver-function | Function for resolving remote paths on the local file-system, and vice versa. | | agent-shell-permission-icon | Icon displayed when shell commands require permission to execute. | | agent-shell-persistent-prompt-enabled | Whether a shell keeps a writable prompt at the buffer end at all times. | | agent-shell-pi-acp-command | Command and parameters for the Pi ACP client. | | agent-shell-pi-environment | Environment variables for the Pi client. | | agent-shell-prefer-viewport-interaction | Non-nil makes ‘agent-shell’ prefer viewport interaction over shell interaction. | | agent-shell-preferred-agent-config | Default agent to use for all new shells. | | agent-shell-prompt-busy-frames | Frames animating the live prompt’s marker while the agent works. | | agent-shell-qoder-acp-command | Command and parameters for the Qoder CLI ACP server. | | agent-shell-qoder-environment | Environment variables for the Qoder CLI ACP server. | | agent-shell-qwen-acp-command | Command and parameters for the Qwen Code client. | | agent-shell-qwen-authentication | Configuration for Qwen Code authentication. | | agent-shell-qwen-environment | Environment variables for the Qwen Code client. | | agent-shell-screenshot-command | The program to use for capturing screenshots. | | agent-shell-section-functions | Abnormal hook run after overlays are applied (experimental). | | agent-shell-session-choices-function | Function to transform the choices offered when starting a shell. | | agent-shell-session-list-page-limit | Maximum number of pages to request when listing sessions. | | agent-shell-session-restore-verbosity | How much prior context to show when restoring a session. | | agent-shell-session-strategy | How to handle sessions when starting a new shell. | | agent-shell-show-busy-indicator | Non-nil to show the busy indicator animation in the header and mode line. | | agent-shell-show-config-icons | Whether to show icons in agent config selection. | | agent-shell-show-context-usage-indicator | Whether and how to show the context usage indicator. | | agent-shell-show-cost-indicator | Whether to show the session’s cumulative cost in the header. | | agent-shell-show-session-id | Non-nil to display the session ID in the header and session selection. | | agent-shell-show-usage-at-turn-end | Whether to display usage information when agent turn ends. | | agent-shell-show-welcome-message | Non-nil to show welcome message. | | agent-shell-status-kind-label-function | Function to render status and kind labels. | | agent-shell-text-file-capabilities | Whether agents are initialized with read/write text file capabilities. | | agent-shell-thought-process-expand-by-default | Whether thought process sections should be expanded by default. | | agent-shell-thought-process-icon | Icon displayed during the AI’s thought process. | | agent-shell-tool-use-expand-by-default | Whether tool use sections should be expanded by default. | | agent-shell-transcript-file-path-function | Function to generate the full transcript file path. | | agent-shell-ui-mode-hook | Hook run after entering or leaving ‘agent-shell-ui-mode’. | | agent-shell-user-message-expand-by-default | Whether user message sections should be expanded by default. | | agent-shell-viewport-dismiss-on-send | Non-nil dismisses the viewport compose window after sending. | | agent-shell-write-inhibit-minor-modes | List of minor mode commands to inhibit during ‘fs/write_text_file’ edits. | | agent-shell-xai-acp-command | Command and parameters for the Grok Build ACP client. | | agent-shell-xai-default-model-id | Default Grok Build model ID. | | agent-shell-xai-default-session-mode-id | Default Grok Build session mode ID. | | agent-shell-xai-environment | Environment variables for the Grok Build ACP client. |

  • Commands

#+BEGIN_SRC emacs-lisp :results table :colnames '("Binding" "Command" "Description") :exports results (let ((rows)) (mapatoms (lambda (symbol) (when (and (string-match "^agent-shell" (symbol-name symbol)) (commandp symbol)) (push `(,(string-join (seq-filter (lambda (symbol) (not (string-match "menu" symbol))) (mapcar (lambda (keys) (key-description keys)) (or (where-is-internal (symbol-function symbol) comint-mode-map nil nil (command-remapping 'comint-next-input)) (where-is-internal symbol agent-shell-mode-map nil nil (command-remapping symbol)) (where-is-internal (symbol-function symbol) agent-shell-mode-map nil nil (command-remapping symbol))))) " or ") ,(symbol-name symbol) ,(car (split-string (or (documentation symbol t) "") "\n"))) rows)))) (sort rows (lambda (a b) (string< (cadr a) (cadr b))))) #+END_SRC

#+RESULTS: | Binding | Command | Description | |-------------------------------------------------------+---------------------------------------------------------+-------------------------------------------------------------------------------| | C-c C-w | agent-shell | Start or reuse an existing agent shell. | | | agent-shell--display-buffer | Toggle agent SHELL-BUFFER display. | | | agent-shell-anthropic-start-claude-code | Start an interactive Claude Agent shell. | | | agent-shell-antigravity-start-agent | Start an interactive Antigravity agent shell. | | | agent-shell-artist-cancel | Discard the scratch art and close the buffer. | | | agent-shell-artist-commit | Insert scratch art into the calling buffer wrapped in a fence. | | | agent-shell-artist-help | Show all interactive commands for agent-shell-artist-mode'. | | | agent-shell-artist-insert-text-drawing | Open the scratch artist buffer to draw and insert text from. | | | agent-shell-artist-mode | Minor mode for agent-shell' scratch artist buffers. | | | agent-shell-artist-next-history | Show newer art from agent-shell-artist-history-ring'. | | | agent-shell-artist-previous-history | Show older art from agent-shell-artist-history-ring'. | | | agent-shell-auggie-start-agent | Start an interactive Auggie agent shell. | | C-M-u | agent-shell-backward-up-item | Go to the start of the item point navigated into. | | | agent-shell-chat-mode | Toggle chat-style Me'/agent labels in the current agent-shell' buffer. | | | agent-shell-clear-buffer | Clear the current shell buffer. | | | agent-shell-cline-start-agent | Start an interactive Cline agent shell. | | | agent-shell-codebuddy-start-agent | Start an interactive CodeBuddy agent shell. | | | agent-shell-completion-mode | Toggle popping completion as @ or / are typed. | | | agent-shell-copy-as-markdown | Copy the region between BEG and END to the kill ring as markdown. | | | agent-shell-copy-last-output | Copy the last agent output to the kill ring. | | | agent-shell-copy-link-url-at-point | Copy the rendered Markdown link URL at POS (or point) to the kill ring. | | | agent-shell-copy-session-id | Copy the current session ID to the kill ring. | | | agent-shell-copy-source-block-at-point | Copy the rendered fenced code block body at POS (or point) to the kill ring. | | | agent-shell-cursor-start-agent | Start an interactive Cursor agent shell. | | C- | agent-shell-cycle-session-mode | Cycle through available session modes for the current agent-shell' session. | | | agent-shell-deepseek-start-agent | Start an interactive DeepSeek Harness agent shell. | | | agent-shell-delete-interaction-at-point | Delete interaction (request and response) at point. | | | agent-shell-diff-accept-all | Accept all changes in the current diff buffer. | | | agent-shell-diff-mode | Major mode for agent-shell' diff buffers. | | | agent-shell-diff-open-file | Open the file for the diff at point and jump to the change. | | | agent-shell-diff-reject-all | Reject all changes in the current diff buffer. | | | agent-shell-droid-start-agent | Start an interactive Factory Droid agent shell. | | | agent-shell-fakes-load-session | Load and replay a recorded ACP session from TRAFFIC-FILE, popping to it. | | | agent-shell-fork | Fork the current session into a new shell. | | | agent-shell-github-start-copilot | Start an interactive GitHub Copilot agent shell. | | | agent-shell-google-start-gemini | Start an interactive Gemini CLI agent shell. | | | agent-shell-goose-start-agent | Start an interactive Goose agent shell. | | | agent-shell-help-menu | Transient menu for agent-shell' commands. | | | agent-shell-hermes-start-agent | Start an interactive Hermes agent shell. | | - | agent-shell-image-scale-decrease | Narrow the image at point, or every image in the buffer, by one step. | | + | agent-shell-image-scale-increase | Widen the image at point, or every image in the buffer, by one step. | | 0 | agent-shell-image-scale-reset | Reset the image at point, or every image in the buffer, to its rendered size. | | | agent-shell-insert-file | Insert a file into agent-shell'. | | | agent-shell-insert-shell-command-output | Execute a shell command and insert output as a code block. | | C-c C-c | agent-shell-interrupt | Interrupt in-progress request and reject all pending permissions. | | | agent-shell-jump-to-latest-permission-button-row | Jump to the latest permission button row. | | | agent-shell-junie-start-agent | Start an interactive Junie agent shell. | | | agent-shell-kimi-start-agent | Start an interactive Kimi Code agent shell. | | | agent-shell-kiro-start-agent | Start an interactive Kiro agent shell. | | | agent-shell-list-edit-dedent-line | De-indent the current list item. | | | agent-shell-list-edit-indent-line | Indent the current list item. | | | agent-shell-list-edit-mode | Minor mode for editing markdown-style lists. | | | agent-shell-list-edit-newline | Insert a newline, continuing the current list item if applicable. | | | agent-shell-markdown-benchmark-compare | Run both renderers ITERATIONS times and print a comparison. | | | agent-shell-markdown-benchmark-stream | Benchmark streaming-render with agent-shell-markdown'. | | | agent-shell-markdown-benchmark-stream-overlays | Benchmark streaming-render with markdown-overlays'. | | | agent-shell-markdown-image-scale-decrease | Narrow the image at point, or every image in the buffer, by one step. | | | agent-shell-markdown-image-scale-increase | Widen the image at point, or every image in the buffer, by one step. | | | agent-shell-markdown-image-scale-reset | Drop sizes set by scaling, restoring the ones the images rendered at. | | | agent-shell-markdown-table-next-cell | Move point to the start of the next table cell. | | | agent-shell-markdown-table-previous-cell | Move point to the start of the previous table cell. | | | agent-shell-mistral-start-vibe | Start an interactive Mistral Vibe agent shell. | | | agent-shell-mock-agent-start-agent | Start an interactive mock ACP agent shell. | | | agent-shell-mode | Major mode for agent shell. | | | agent-shell-narrow-to-block | Narrow to the last COUNT navigatable blocks in the current buffer. | | | agent-shell-new-downloads-shell | Start a new agent shell in ~/Downloads. | | | agent-shell-new-shell | Start a new agent shell. | | | agent-shell-new-temp-shell | Start a new agent shell in a temporary directory. | | | agent-shell-new-worktree-shell | Create a new git worktree and start an agent shell in it. | | S- | agent-shell-newline | Insert a newline, and move to left margin of the new line. | | C- or M-n | agent-shell-next-input | Cycle forwards through input history. | | n or TAB | agent-shell-next-item | Go to next item. | | | agent-shell-next-permission-button | Jump to the next button. | | | agent-shell-omp-start | Start an interactive Oh My Pi (omp) agent shell. | | | agent-shell-open-transcript | Open the transcript file for the current agent-shell' buffer. | | | agent-shell-openai-start-codex | Start an interactive Codex agent shell. | | | agent-shell-openclaw-start-agent | Start an interactive OpenClaw agent shell. | | | agent-shell-opencode-start-agent | Start an interactive OpenCode agent shell. | | C-c C-o | agent-shell-other-buffer | Switch to other associated buffer (viewport vs shell). | | | agent-shell-pi-start-agent | Start an interactive Pi coding agent shell. | | C-<up> or M-p | agent-shell-previous-input | Cycle backwards through input history, saving input. | | p or <backtab> | agent-shell-previous-item | Go to previous item. | | | agent-shell-previous-permission-button | Jump to the previous button. | | | agent-shell-prompt-compose | Compose an agent-shell' prompt in a dedicated buffer. | | | agent-shell-prompt-queue | Queue or immediately send a prompt depending on shell busy state. | | | agent-shell-prompt-queue-remove | Remove all pending prompts or a specific prompt by REMOVE-INDEX. | | | agent-shell-prompt-queue-resume | Resume processing pending prompts in the queue. | | | agent-shell-prompt-steer | Steer PROMPT into the turn the agent is currently running. | | | agent-shell-qoder-start-agent | Start an interactive Qoder agent shell. | | r | agent-shell-quote-region | Quote the active region into the shell's latest prompt. | | | agent-shell-qwen-start | Start an interactive Qwen Code CLI agent shell. | | | agent-shell-reload | Reload the current session by restarting with the same session ID. | | C-x x r | agent-shell-rename-buffer | Rename current shell buffer. | | | agent-shell-reset-logs | Reset all log buffers. | | | agent-shell-restart | Clear conversation by restarting the agent shell in the same project. | | | agent-shell-resume-session | Resume an existing agent session by SESSION-ID. | | | agent-shell-run-all-tests | Run all agent-shell tests in batch mode. | | | agent-shell-save-traffic | Save agent shell traffic to a .traffic file and reveal it in dired'. | | M-r | agent-shell-search-history | Search previous input history. | | | agent-shell-send-clipboard-image | Paste clipboard image and insert it into agent-shell'. | | | agent-shell-send-clipboard-image-to | Like agent-shell-send-clipboard-image' but prompt for which shell to use. | | | agent-shell-send-current-file | Insert a file into agent-shell'. | | | agent-shell-send-dwim | Send region or error at point to last accessed shell buffer in project. | | | agent-shell-send-file | Insert a file into agent-shell'. | | | agent-shell-send-file-to | Like agent-shell-send-file' but prompt for which shell to use. | | | agent-shell-send-other-file | Prompt to send a file into agent-shell'. | | | agent-shell-send-region | Send region to last accessed shell buffer in project. | | | agent-shell-send-region-to | Like agent-shell-send-region' but prompt for which shell to use. | | | agent-shell-send-screenshot | Capture a screenshot and insert it into agent-shell'. | | | agent-shell-send-screenshot-to | Like agent-shell-send-screenshot' but prompt for which shell to use. | | C-c C-s | agent-shell-set-session-config-option | Set a session config option. | | C-c RET | agent-shell-set-session-mode | Set session mode (if any available). | | C-c C-v | agent-shell-set-session-model | Set session model. | | C-c C-t | agent-shell-set-session-thought-level | Set thought level (reasoning effort) for the current session. | | | agent-shell-show-usage | Display current session usage information in the minibuffer. | | RET | agent-shell-submit | Submit the current input to the agent, or queue it while the agent is busy. | | M- or M-RET | agent-shell-submit-override | Submit the current input through the override route. | | | agent-shell-switch-buffer | Switch to another agent-shell-mode' buffer via completing-read'. | | | agent-shell-toggle | Toggle agent shell display. | | | agent-shell-toggle-logging | Toggle logging. | | | agent-shell-ui-backward-block | Jump to the previous block. | | | agent-shell-ui-forward-block | Jump to the next block. | | | agent-shell-ui-mode | Minor mode for SUI block navigation. | | | agent-shell-ui-toggle-all-fragments | Toggle global fold state: all-expanded ↔ all-collapsed. | | | agent-shell-ui-toggle-fragment | Toggle fragment fold at or near point. | | | agent-shell-version | Show agent-shell' mode version. | | | agent-shell-view-acp-logs | View agent shell ACP logs buffer. | | | agent-shell-view-traffic | View agent shell traffic buffer. | | | agent-shell-viewport--compose-help-menu | agent-shell' viewport compose help menu. | | | agent-shell-viewport--help-menu | agent-shell' viewport help menu. | | | agent-shell-viewport-compose-cancel | Cancel prompt composition. | | | agent-shell-viewport-compose-help-menu | Show the transient help menu for compose (edit) mode. | | | agent-shell-viewport-compose-peek-last | Save compose buffer snapshot and peek at the last interaction. | | | agent-shell-viewport-compose-send | Send the viewport composed prompt to the agent shell. | | | agent-shell-viewport-compose-send-and-dismiss | Queue or send the composed prompt, then dismiss the compose window. | | | agent-shell-viewport-compose-send-and-kill | Send the viewport composed prompt to the agent shell and kill compose buffer. | | | agent-shell-viewport-compose-send-and-wait-for-response | Send the viewport composed prompt and display response in viewport. | | | agent-shell-viewport-compose-send-override | Send the viewport composed prompt through the override route. | | | agent-shell-viewport-copy-session-id | Copy the current session ID to the kill ring. | | | agent-shell-viewport-cycle-session-mode | Cycle through available session modes. | | | agent-shell-viewport-edit-mode | Major mode for composing agent shell prompts. | | | agent-shell-viewport-help-menu | Show viewport and display the transient help menu (bound to ? in view mode). | | | agent-shell-viewport-interrupt | Interrupt active agent shell request. | | | agent-shell-viewport-next-history | Insert next prompt from history into compose buffer. | | | agent-shell-viewport-next-item | Go to next item. | | | agent-shell-viewport-next-page | Show next interaction (request / response). | | | agent-shell-viewport-open-transcript | Open the transcript file for the current agent-shell' session. | | | agent-shell-viewport-previous-history | Insert previous prompt from history into compose buffer. | | | agent-shell-viewport-previous-item | Go to previous item. | | | agent-shell-viewport-previous-page | Show previous interaction (request / response). | | | agent-shell-viewport-prompt-queue | Queue or immediately send a prompt depending on shell busy state. | | | agent-shell-viewport-prompt-queue-remove | Remove pending prompts. | | | agent-shell-viewport-prompt-queue-resume | Resume processing pending prompts in the queue. | | | agent-shell-viewport-quote-reply | Reply with the entire response block-quoted. | | | agent-shell-viewport-refresh | Refresh viewport buffer content with current item from shell. | | | agent-shell-viewport-reply | Reply as a follow-up and compose another prompt/query. | | | agent-shell-viewport-reply-1 | Reply with "1" and send immediately. | | | agent-shell-viewport-reply-2 | Reply with "2" and send immediately. | | | agent-shell-viewport-reply-3 | Reply with "3" and send immediately. | | | agent-shell-viewport-reply-4 | Reply with "4" and send immediately. | | | agent-shell-viewport-reply-5 | Reply with "5" and send immediately. | | | agent-shell-viewport-reply-6 | Reply with "6" and send immediately. | | | agent-shell-viewport-reply-7 | Reply with "7" and send immediately. | | | agent-shell-viewport-reply-8 | Reply with "8" and send immediately. | | | agent-shell-viewport-reply-9 | Reply with "9" and send immediately. | | | agent-shell-viewport-reply-again | Reply with "again" and send immediately. | | | agent-shell-viewport-reply-continue | Reply with "continue" and send immediately. | | | agent-shell-viewport-reply-more | Reply with "more" and send immediately. | | | agent-shell-viewport-reply-yes | Reply with "yes" and send immediately. | | | agent-shell-viewport-save-traffic | Save agent shell traffic to a .traffic file and reveal it in dired'. | | | agent-shell-viewport-search-history | Search prompt history, select, and insert into compose buffer. | | | agent-shell-viewport-set-session-mode | Set session mode. | | | agent-shell-viewport-set-session-model | Set session model. | | | agent-shell-viewport-set-session-thought-level | Set thought level (reasoning effort) for the current session. | | | agent-shell-viewport-view-acp-logs | View agent shell ACP logs buffer. | | | agent-shell-viewport-view-last | Display the last request/response interaction. | | | agent-shell-viewport-view-mode | Major mode for viewing agent shell prompts (read-only). | | | agent-shell-viewport-view-traffic | View agent shell traffic buffer. | | | agent-shell-xai-start-grok | Start an interactive Grok Build agent shell. | | C-y or s-v or <paste> or S-<insertchar> or S-<insert> | agent-shell-yank-dwim | Yank or paste clipboard image into agent-shell'. |

Please read through this section before filing issues or feature requests. I won't be able to help unless I have enough information to help. This section shows how to get the details I need.

  • Versions: What version of =agent-shell=, =acp.el=, the ACP package (e.g. =claude-code-acp=), and the agent CLI (e.g. =claude=, =gemini=) are you running?
  • Steps to reproduce: What exact steps lead to the issue? Be specific about what you did and in what order.
  • ACP traffic: See [[https://github.com/xenodium/agent-shell?tab=readme-ov-file#how-do-i-viewget-agent-client-protocol-traffic][how to get ACP traffic]] above. This is the most useful piece of information for diagnosing issues.
  • Screenshots: A screenshot helps correlate what you see with the protocol data.
  • Your agent-shell config: Share any relevant =agent-shell= variable settings from your Emacs config.
  • Profiling data (for performance issues): Use =M-x profiler-start=, reproduce the issue, then =M-x profiler-report= (and =M-x profiler-stop=). Share the report.
  • FAQ

** Why doesn't =agent-shell= offer all slash commands/skills available in CLI agent?

=agent-shell= can only offer the slash commands/skills advertised by the agent via [[https://agentclientprotocol.com][Agent Client Protocol]]. To view what's exposed by your agent, expand the "Available /commands" section. Is the command you're after missing? Please consider filing a feature-request with the respective agent (ie. Gemini CLI) or their ACP layer (claude-code-acp).

[[file:slash-commands.png]]

** Can you add support for another agent?

Does the agent support ACP ([[https://agentclientprotocol.com][Agent Client Protocol]])? If so, =agent-shell= can likely support this agent. Some agents have ACP support built-in (like [[https://github.com/google-gemini/gemini-cli][gemini-cli]]). Others require a separate ACP package (like [[https://github.com/zed-industries/claude-code-acp][claude-code-acp]] for [[https://github.com/anthropics/claude-code][claude-code]]). When filing a feature request to add a new agent, please include a link to the project supporting [[https://agentclientprotocol.com][Agent Client Protocol]] (built-in or otherwise).

Agents without ACP support are out of scope for integrating with =agent-shell=. Having said that, if you do build an ACP layer like =claude-agent-acp=, then =agent-shell= can work with it.

** =agent-shell= not behaving as expected?

Could be the agent itself missing an [[https://agentclientprotocol.com][Agent Client Protocol]] feature, =agent-shell= missing the feature, or both :) So which one is it? It's hard to tell unless we look at [[https://agentclientprotocol.com][Agent Client Protocol]] traffic between the two.

** Is shx supported? No. There are reports that [[https://github.com/xenodium/agent-shell/issues/543][it does not work]] with agent-shell. ** Agent hangs at "Loading..." or fails with "Method not found: initialize" over TRAMP?

When =default-directory= is remote, the agent is started on the remote host via TRAMP. By default, TRAMP starts it through its own remote shell, which was reported to hang at "Loading..." in [[https://github.com/xenodium/agent-shell/issues/872][#872]]. Enabling TRAMP's direct asynchronous processes for the host (along with the =direct-async= setting below) resolved it:

#+begin_src emacs-lisp (connection-local-set-profile-variables 'remote-direct-async-process '((tramp-direct-async-process . t)))

(connection-local-set-profiles '(:application tramp :machine "my-host") 'remote-direct-async-process) #+end_src

For the =ssh=, =scp=, and =rsync= methods, TRAMP runs direct asynchronous processes with =ssh -t -t=, which allocates a remote terminal. In #872, that terminal echoed =agent-shell='s requests back, failing with "Method not found: initialize". The following tells TRAMP to skip =-t -t= for the host:

#+begin_src emacs-lisp (add-to-list 'tramp-connection-properties (list (regexp-quote "/ssh:my-host:") "direct-async" t)) #+end_src

Direct asynchronous processes don't support interactive authentication (e.g. password prompts), and TRAMP falls back to its default implementation for multi-hop connections. See TRAMP's "Improving performance of asynchronous remote processes" manual section for details.

Agent executables are looked up on the remote host via =tramp-remote-path= (not =exec-path=). If yours isn't found, add its remote directory to =tramp-remote-path= and run =M-x tramp-cleanup-connection=, as TRAMP caches the remote path per connection.

For mapping remote file paths between Emacs and the agent, see [[https://github.com/junyi-hou/agent-shell-tramp][agent-shell-tramp]]. ** How do I view/get [[https://agentclientprotocol.com][Agent Client Protocol]] traffic?

  1. =M-x agent-shell-toggle-logging= (make sure logging is ON).
  2. Reproduce the issue
  3. =M-x agent-shell-view-traffic=

Browse through traffic and see if you can spot the issue. For example, if you see a request sent by the agent asking for user permission, but =agent-shell= isn't surfacing this permission, it looks like perhaps =agent-shell= is missing a feature.

For example, here's what a [[https://agentclientprotocol.com/protocol/schema#session%2Frequest-permission][session/request_permission]] request would look like from the traffic viewer.

[[file:traffic.png]]

Sometimes including a traffic screenshot in an issue is enough. Other times including the full traffic is needed. From the shell or viewport, you can =M-x agent-shell-save-traffic= (or =M-x agent-shell-viewport-save-traffic=) to save as =.traffic=. From the traffic viewer itself, =M-x acp-traffic-save-to= does the same.

** Where should I file bug or feature request?

*** Agent issues or feature requests

If you're able to determine the agent is missing a feature (or a bug is present) in their [[https://agentclientprotocol.com][Agent Client Protocol]] implementation, please file an issue directly with the agent folks. For example:

*** =agent-shell= issues or feature requests

Alternatively, if you noticed =agent-shell= is missing a feature (or has a bug), please [[https://github.com/xenodium/agent-shell/issues][file an agent-shell issue]].

*** Not sure where to file an issue?

File [[https://github.com/xenodium/agent-shell/issues][in agent-shell]], but please try to provide details, so I can determine whether =agent-shell= or the agent itself needs work. Traffic data would be very useful here. Provide a screenshot or a .traffic file if you think it'll help. See [[https://github.com/xenodium/agent-shell?tab=readme-ov-file#how-do-i-viewget-agent-client-protocol-traffic][how to get ACP traffic]].

  • Contributing

See [[file:CONTRIBUTING.org][CONTRIBUTING.org]].

  • Contributors

#+HTML: #+HTML: #+HTML:

Made with [[https://contrib.rocks][contrib.rocks]].

Explore related projects

affaan-m/ECC

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

299,095 stars

NousResearch/hermes-agent

The agent that grows with you

261,465 stars

multica-ai/andrej-karpathy-skills

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

230,590 stars

firecrawl/firecrawl

Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥

199,303 stars