IBM Paver chat commands reference
The IBM® Paver chat interface provides slash commands and controls for managing interactive sessions in both the terminal and web UI. Use chat commands to navigate conversation history, switch working contexts, manage keyring secrets, and control sandbox execution.
On this page
For paver CLI commands, see the IBM Paver CLI command reference.
Start a chat session#
In the web UI, click New task to start a new chat. If you're using the CLI, use the paver chat command.
Slash commands#
Slash commands are typed directly into the paver chat input. They control the session without sending a message to the agent.
| Command | Description |
|---|---|
/clear, /cls | Clears the chat screen without resetting the conversation history or session context. Use this to clear visual clutter in your terminal. |
/context | View and switch between working contexts (IBM Cloud® accounts). Use this when you need Paver to work across accounts mid-conversation. For example, you might have an application's backups live in a different account than the application itself. Subcommands include current to view the active context, list to show all available contexts, switch <name> to switch to a specific context, and back to return to the previous context. Conversations record which context they were in. If you resume a conversation from a different context, Paver asks whether you want to switch to the conversation's original context or move the conversation to your current one. |
/conv | Manage conversation history. Subcommands include list for viewing previous conversations and their IDs, resume <number> for restoring and continuing a past conversation by ID, and clear for deleting saved conversation history. |
/edit | Switch the safety mode to Edit mode, which allows Paver to make changes in your environment subject to your approval. |
/exit, /quit, /q | Exit the interactive chat session and return to your terminal. |
/expand | Control the level of detail displayed in the agent's work section during task execution. Options include default for compact progress ticker and verbose for expand and collapse detailed tool calls and agent reasoning. |
/feedback | Submit feedback and diagnostic logs for agent interactions. Use /feedback create <text> to submit feedback in the web UI, /feedback <text> in an interactive CLI chat, and /feedback list to view submitted feedback. |
/files | Manage files in the remote sandbox workspace. Subcommands include list to view workspace files, upload <path> to upload local files or folders into the sandbox, download <filename> to download sandbox files or folders to your machine, remove <filename> to delete files, and clear to remove all workspace files. |
/guard | Inspect or configure Paver Guard safety policies for the current session. Subcommands include status to view active policy, reset to restore default guard rules (auto-approve-low), or applying a specific policy name (such as auto-approve-low, auto-approve-medium, auto-deny-medium, auto-deny-high, or todo <smart|true|false>). |
/help, /? | Display the list of available slash commands and usage help directly in the chat session. |
/info | Display technical details about the active conversation and session. |
/keyring | Manage encrypted secrets that are injected into the sandbox as environment variables when exported. Subcommands include list to view stored secrets and expiration, add <name> <source> (store a secret from env://VAR_NAME or file://path), export to export secrets into the sandbox shell, remove <name> to delete a stored secret, and clear to remove all secrets. |
/lang | Configure language preferences for the chat interface and agent responses. Subcommands include show to view active language settings, list to display supported language tags, <tag> to set both interface and response language, response <tag> to set response language only, interface <tag> to set UI language only, and reset to restore default language. |
/login | Reauthenticate your Paver session without restarting the chat. Options include apikey for logging in using an API key prompt and sso to authenticate by using single sign-on. Running without arguments uses the API key in your .env file if available. |
/memory | Manage persistent context and facts Paver remembers about you across sessions. Subcommands include list to view active memories, add <text> to manually save a memory, full to view complete memory details, remove <id> to delete a specific memory, clear to delete all memories, and export or import to back up or restore memories to or from a file. |
/new | Start a brand-new conversation in the current session. Clears active conversation context while retaining persistent memories, keyring secrets, and sandbox workspace files. |
/personalize | In the web UI, start a guided interview that builds your Paver memory profile. Paver asks a series of questions about your role, preferences, and environment, then saves the answers as persistent memories. |
/safe | Switch safety mode to Safe mode, which prevents Paver from making changes in your environment. Paver can still read your environment and read and write files in the sandbox. |
/shell | Open an interactive shell inside the sandbox container (press Ctrl+] to return to chat), or manage shell host bindings for external infrastructure. Subcommands include current to view active host, list to list registered hosts, register <name> to add a host binding, use <name> to switch the active host, and remove <name> to delete a host binding. |
/shortcut | Create, manage, and run reusable prompt templates and shortcuts. Subcommands include list to view saved shortcuts, save <name> <prompt> to save a new shortcut, show <name> to display a shortcut's prompt template, remove <name> to delete a shortcut, export or import to save or load shortcuts to or from a file, and help to show shortcut usage help. |
/skill | Manage custom skills for extending Paver's expertise with your workflows and knowledge. Subcommands include /skill list to list all uploaded skills, /skill upload <path> to upload a skill, and /skill remove <name> to remove a custom skill. |
/skills | Manage the custom skills available in your session. Subcommands include list to view uploaded skills, upload <path> to upload and install a new custom skill package, and remove <name> to delete a custom skill. |
/theme | Customize the terminal UI appearance and progress-indicator color palette. Subcommands include show to view the active theme, list to display available themes, update <name> to apply a theme, reset to restore default theme, prefer [dark|light] to set dark or light preference, and auto-switch [on|off] to toggle automatic theme adaptation. |
/timers | Toggle elapsed-time display the elapsed time for each individual tool call or task step it performs while working toward your goal. Use on to show running durations for each task step or off to hide timers. |
/todo | Display the agent's current step-by-step task checklist and track completion progress for the active task. |
Note
Available slash commands might vary based on where you're using Paver. Use the /help command in an interactive chat session to see the full list of supported commands.
Direct bash#
Prefix any input with bash or ! to run a command directly in the sandbox, bypassing the agent and the approval system:
bash kubectl get pods -n production
! ibmcloud ks cluster lsDirect bash commands execute immediately without an approval prompt. Use this when you know exactly what you want to run and don't need Paver to reason about it.
Session startup flags#
These flags are passed to paver chat when launching the session:
| Flag | Description |
|---|---|
--approve-execute-bash | Auto-approve shell command tool calls only |
--approve-read-file | Auto-approve file read tool calls only |
--approve-write-file | Auto-approve file write tool calls only |
--verbose | Show tool calls and agent reasoning as they happen |
-c, --conversation <id> | Continue an existing conversation by ID |
--guard <policy> | Set the session-scoped Paver Guard policy |