Ai Keeper/Guides/Chat and Assistant
← All workflows
Collection 03 · Everyday use

Chat and Assistant

Start conversations, attach files, use tools with approvals, manage long sessions, export work, and use the dedicated or floating assistant.

73 workflowsBeginner-first

Find one workflow

Filter this collection by name, outcome, provider, engine, channel, mode, or command.

73 shown
Workflow 01

Start a clean conversation

Pick a Ready model, send a clear request, and keep the conversation easy to find.

Beginner2 minResult · A saved conversation streams a reply✓ Source-audited workflow
Clean Chat start screen with no conversations yet, one available model, Tools capability, and the message composer.
A blank Chat makes the order clear: choose a model, optionally adjust its controls, then type the first message.
Before you start
  • At least one Ready Chat/LLM instance or configured provider.

  1. Open Chat and click the new-conversation control.

  2. Choose the model by its model-scoped label. Prefer a smaller fast model for simple tasks and a stronger model for complex reasoning.

  3. Write the goal, useful context, desired output, and any constraints in one message. Then send.

  4. Rename or place the conversation in a sidebar folder if you expect to return to it.

You are done when
Text arrives progressively, the model label remains the one you chose, and the conversation appears in the sidebar.
If something looks wrong
Send button does nothingConfirm the selected model is Ready and the composer is not waiting on an approval.
Answer is vagueState the intended result and format; do not compensate by changing ten model settings.
Workflow 02

Ask about a file or image

Attach temporary material to one conversation without adding it to permanent knowledge.

Beginner3–10 minResult · The answer refers to the attachment✓ Source-audited workflow
Clean Chat start screen with the message composer and attachment controls.
Start the conversation first; the plus control beside the composer is where files and images are attached.
Before you start
  • A readable local file.
  • For images, a vision-capable model or route.

  1. Open Chat and use the attachment control to select the file or image.

  2. Wait until the attachment is visibly listed in the composer. Remove any accidental file before sending.

  3. Ask a specific question such as Summarize this PDF in five bullets and cite the page for each bullet.

  4. For an image, confirm the selected model supports vision. For repeated future use, index the source under Knowledge instead of attaching it every time.

You are done when
The answer discusses details that are present in the attachment, and any tool/attachment status finishes without error.
If something looks wrong
The model says it cannot see imagesSwitch to a vision-capable instance; text-only models cannot infer pixels.
Large document overwhelms the chatUse Knowledge > Documents for retrieval instead of placing the entire file in one request.
Workflow 03

Let Chat use tools safely

Enable only the tools needed for the task and review sensitive calls before they run.

Beginner5 minResult · A tool call completes and the model uses its result✓ Source-audited workflow
Clean Chat start screen with an empty conversation list, the available model, Tools capability, and message composer.
Choose the model first; tool choices and their approvals become part of the conversation after it starts.
Before you start
  • A model or external route with tool-calling enabled.
  • At least one relevant built-in or MCP tool.

  1. In Chat, open the tools control or type /tools to inspect what is available.

  2. Enable the smallest relevant set. Do not grant shell, browser, file-write, or messaging tools for a question that only needs date/time.

  3. Ask for an action that clearly needs one enabled tool.

  4. When an approval card appears, read the tool name and arguments. Approve only the exact intended action; otherwise reject and explain the safer alternative.

  5. Wait for the assistant's follow-up after the tool result. A raw tool result alone is not the finished answer.

You are done when
The transcript contains assistant tool call, tool result, and a final assistant response; System > Advanced > Audit Trail records the action.
If something looks wrong
The model repeats the same toolStop the turn and inspect the tool error. Rewording the task will not fix a missing permission or malformed argument.
Too many tools exceed contextUse a smaller relevant tool group; self-hosted runtimes receive a reduced catalog for this reason.
Workflow 04

Use slash commands

Discover and run built-in session, context, tool, agent, approval, and export actions from the composer.

Beginner3 minResult · A command changes or reports chat state✓ Source-audited workflow
Clean Chat start screen whose composer accepts slash commands after a model is selected.
Type a slash in the composer to open live command completion instead of memorizing syntax.
Before you start
  • A Chat conversation is open.

  1. Type / in the composer and pause. Choose from the suggestion menu rather than guessing a command name.

  2. Use /commands for the current full list, /status for runtime state, /context for context use, and /tools for tools.

  3. Use commands such as /compact, /export, /reset, or /stop only after reading their description in the suggestion row.

  4. Confirm the report or action shown in the transcript.

You are done when
The command is recognized as an action/report rather than sent as ordinary prompt text.
If something looks wrong
A copied command is unknownUse /commands. Old documentation or custom skills may mention commands that are not registered.
Used /vc expecting version control/vc controls voice-channel behavior; it is not a Git command.
Workflow 05

Keep a long conversation going

Check context use and compact older messages before the model runs out of room.

Beginner3–10 minResult · The chat continues with a smaller working history✓ Source-audited workflow
Clean Chat screen showing the conversation list and the main conversation workspace.
Compaction belongs to an active conversation; use a new disposable chat when learning the controls.
Before you start
  • A conversation with useful earlier context.

  1. Type /context or inspect the context indicator to see how full the model window is.

  2. Before it reaches the limit, type /compact. Let the summarization finish.

  3. Read the resulting summary. Correct any critical fact that was compressed incorrectly.

  4. Continue with a short recap of the immediate next goal. If the topic has completely changed, start a new conversation instead.

You are done when
Context use drops, a compacted summary is present, and the next response still respects the important goals and constraints.
If something looks wrong
Compaction cannot startMake sure a summarizer-capable route is available and the current turn is not still streaming.
Old detail was lostKeep durable facts in Knowledge and source material in Documents; compaction is a summary, not an archive replacement.
Workflow 06

Regenerate, fork, export, import, or archive a session

Preserve useful work and branch experiments without damaging the original conversation.

Beginner3–10 minResult · The desired session copy or file exists✓ Source-audited workflow
Clean Chat screen with conversation-list management controls and an empty first-use state.
New Chat, folders, search, and conversation actions stay together in the left pane.
Before you start
  • An existing conversation.

  1. Use Regenerate when you want a different answer to the same last prompt.

  2. Use the fork action when you want to explore a different direction while preserving the original history.

  3. Use /export or the session export control to save a portable record. Choose the offered format and destination.

  4. Use the import action for a supported Ai Keeper session file; inspect it before continuing tool-enabled work.

  5. Open System > Sessions to archive, restore, or audit older sessions.

You are done when
The original remains intact, the fork or regenerated branch is distinct, or the exported file opens and contains the expected transcript.
If something looks wrong
Export contains sensitive dataStore it like any other private document; attachments, tool arguments, and model output may be included.
Imported session requests toolsReview the current tool policy. Imported text does not inherit trust.
Workflow 07

Use the dedicated or floating Assistant

Keep quick help available without mixing it into an audited multi-agent workspace.

Beginner3 minResult · Assistant answers in-app or in its floating window✓ Source-audited workflow
Current Assistant hub with enablement, floating-window state, and quick actions.
The Assistant surface shows whether it is enabled, visible, detached, and ready for a one-turn test.
Before you start
  • A Ready conversational model or provider route.

  1. Open Assistant and choose the model/profile it should use.

  2. Send a short request and confirm the dedicated surface answers.

  3. Use the detach or floating-window control when you want the assistant to remain above other work.

  4. Return to Chat for carefully managed sessions or Workspaces for structured multi-agent runs.

You are done when
The Assistant responds, and detaching creates a usable floating presence without closing the main app.
If something looks wrong
Floating window is not a workspaceIt is a convenience surface. Use Workspaces when roles, stages, reports, and approval trace matter.
Workflow 08

Use /summarize

Summarize the conversation so far. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /summarize produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /summarize, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /summarize. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/summarize returns a visible report, state change, export/import result, or controlled action matching “Summarize the conversation so far”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 09

Use /translate

Translate text to another language. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /translate produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /translate, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /translate to German: Hello, how are you?. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/translate returns a visible report, state change, export/import result, or controlled action matching “Translate text to another language”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 10

Use /code

Ask for code generation. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /code produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /code, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /code Write a Swift function that adds two integers.. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/code returns a visible report, state change, export/import result, or controlled action matching “Ask for code generation”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 11

Use /explain

Explain a concept or code. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /explain produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /explain, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /explain What is retrieval-augmented generation?. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/explain returns a visible report, state change, export/import result, or controlled action matching “Explain a concept or code”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 12

Use /fix

Fix bugs in code. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /fix produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /fix, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /fix Explain the bug in this pasted snippet before changing it.. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/fix returns a visible report, state change, export/import result, or controlled action matching “Fix bugs in code”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 13

Use /review

Review code for improvements. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /review produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /review, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /review Review this pasted function for correctness.. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/review returns a visible report, state change, export/import result, or controlled action matching “Review code for improvements”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 14

Use /brainstorm

Brainstorm ideas. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /brainstorm produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /brainstorm, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /brainstorm Five names for a local AI project.. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/brainstorm returns a visible report, state change, export/import result, or controlled action matching “Brainstorm ideas”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 15

Use /proscons

List pros and cons. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /proscons produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /proscons, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /proscons Local model versus cloud model.. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/proscons returns a visible report, state change, export/import result, or controlled action matching “List pros and cons”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 16

Use /memory

Show what the assistant remembers about you. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /memory produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /memory, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /memory. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/memory returns a visible report, state change, export/import result, or controlled action matching “Show what the assistant remembers about you”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 17

Use /clear

Start a new conversation. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /clear produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /clear only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /clear. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/clear returns a visible report, state change, export/import result, or controlled action matching “Start a new conversation”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 18

Use /regenerate

Regenerate the last response. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /regenerate produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /regenerate, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /regenerate. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/regenerate returns a visible report, state change, export/import result, or controlled action matching “Regenerate the last response”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 19

Use /agent

Toggle agent mode for multi-step tasks. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /agent produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /agent, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /agent. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/agent returns a visible report, state change, export/import result, or controlled action matching “Toggle agent mode for multi-step tasks”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 20

Use /compact

Summarize and trim the conversation. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /compact produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /compact, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /compact. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/compact returns a visible report, state change, export/import result, or controlled action matching “Summarize and trim the conversation”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 21

Use /status

Show active model, tasks, and system status. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /status produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /status, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /status. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/status returns a visible report, state change, export/import result, or controlled action matching “Show active model, tasks, and system status”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 22

Use /context

Show effective prompt and active policies for current agent. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /context produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /context, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /context. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/context returns a visible report, state change, export/import result, or controlled action matching “Show effective prompt and active policies for current agent”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 23

Use /plugins

Show installed ClawHub packages. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /plugins produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /plugins, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /plugins. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/plugins returns a visible report, state change, export/import result, or controlled action matching “Show installed ClawHub packages”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 24

Use /mcp

Show connected MCP servers. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /mcp produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /mcp, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /mcp. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/mcp returns a visible report, state change, export/import result, or controlled action matching “Show connected MCP servers”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 25

Use /models

Show available and running models. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /models produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /models, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /models. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/models returns a visible report, state change, export/import result, or controlled action matching “Show available and running models”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 26

Use /tasks

Show scheduled tasks. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /tasks produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /tasks, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /tasks. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/tasks returns a visible report, state change, export/import result, or controlled action matching “Show scheduled tasks”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 27

Use /hooks

Show active hooks. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /hooks produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /hooks, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /hooks. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/hooks returns a visible report, state change, export/import result, or controlled action matching “Show active hooks”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 28

Use /orders

Show standing orders. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /orders produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /orders, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /orders. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/orders returns a visible report, state change, export/import result, or controlled action matching “Show standing orders”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 29

Use /find

Search Skills + MCP registries for a term. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /find produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /find, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /find calendar. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/find returns a visible report, state change, export/import result, or controlled action matching “Search Skills + MCP registries for a term”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 30

Use /snapshot

Snapshot a directory before risky tool runs. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /snapshot produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /snapshot only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /snapshot /path/to/project -- before update. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/snapshot returns a visible report, state change, export/import result, or controlled action matching “Snapshot a directory before risky tool runs”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 31

Use /rollback

Restore the most recent snapshot. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /rollback produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /rollback only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /rollback <snapshot-id>. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/rollback returns a visible report, state change, export/import result, or controlled action matching “Restore the most recent snapshot”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 32

Use /recommend

Surface capability gaps in the active toolset. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /recommend produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /recommend, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /recommend. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/recommend returns a visible report, state change, export/import result, or controlled action matching “Surface capability gaps in the active toolset”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 33

Use /commitments

List inferred follow-up commitments. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /commitments produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /commitments, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /commitments. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/commitments returns a visible report, state change, export/import result, or controlled action matching “List inferred follow-up commitments”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 34

Use /commit

Add a follow-up commitment manually. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /commit produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /commit only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /commit 2h review the authentication notes. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/commit returns a visible report, state change, export/import result, or controlled action matching “Add a follow-up commitment manually”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 35

Use /ingest

Index a local file or folder for document_search. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /ingest produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /ingest only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /ingest /path/to/document.pdf. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/ingest returns a visible report, state change, export/import result, or controlled action matching “Index a local file or folder for document_search”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 36

Use /ingested

List ingested documents. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /ingested produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /ingested, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /ingested. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/ingested returns a visible report, state change, export/import result, or controlled action matching “List ingested documents”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 37

Use /forget

Drop ingested documents matching a pattern. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /forget produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /forget only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /forget document.pdf. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/forget returns a visible report, state change, export/import result, or controlled action matching “Drop ingested documents matching a pattern”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 38

Use /watch

Watch a folder and auto-reingest files on change. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /watch produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /watch only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /watch /path/to/notes. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/watch returns a visible report, state change, export/import result, or controlled action matching “Watch a folder and auto-reingest files on change”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 39

Use /unwatch

Stop watching a folder. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /unwatch produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /unwatch only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /unwatch /path/to/notes. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/unwatch returns a visible report, state change, export/import result, or controlled action matching “Stop watching a folder”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 40

Use /preset

Save, apply, list, or delete per-chat config presets. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /preset produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /preset only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /preset list. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/preset returns a visible report, state change, export/import result, or controlled action matching “Save, apply, list, or delete per-chat config presets”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 41

Use /verbose

Ask for a detailed response. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /verbose produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /verbose, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /verbose. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/verbose returns a visible report, state change, export/import result, or controlled action matching “Ask for a detailed response”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 42

Use /reasoning

Show step-by-step reasoning. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /reasoning produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /reasoning, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /reasoning. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/reasoning returns a visible report, state change, export/import result, or controlled action matching “Show step-by-step reasoning”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 43

Use /exec

Configure exec policy or run an allowed shell command. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /exec produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /exec only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /exec status. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/exec returns a visible report, state change, export/import result, or controlled action matching “Configure exec policy or run an allowed shell command”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 44

Use /new

Start a new conversation. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /new produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /new only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /new. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/new returns a visible report, state change, export/import result, or controlled action matching “Start a new conversation”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 45

Use /reset

Reset agent session context. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /reset produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /reset only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /reset. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/reset returns a visible report, state change, export/import result, or controlled action matching “Reset agent session context”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 46

Use /stop

Stop current generation. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /stop produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /stop, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /stop. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/stop returns a visible report, state change, export/import result, or controlled action matching “Stop current generation”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 47

Use /session

Show or edit current session and thread bindings. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /session produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /session, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /session. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/session returns a visible report, state change, export/import result, or controlled action matching “Show or edit current session and thread bindings”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 48

Use /tools

Show available tools for the current agent. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /tools produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /tools, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /tools. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/tools returns a visible report, state change, export/import result, or controlled action matching “Show available tools for the current agent”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 49

Use /commands

Show all available slash commands. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /commands produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /commands, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /commands. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/commands returns a visible report, state change, export/import result, or controlled action matching “Show all available slash commands”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 50

Use /agents

Show configured agents and active agent sessions. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /agents produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /agents, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /agents. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/agents returns a visible report, state change, export/import result, or controlled action matching “Show configured agents and active agent sessions”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 51

Use /think

Toggle step-by-step reasoning mode. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /think produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /think, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /think. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/think returns a visible report, state change, export/import result, or controlled action matching “Toggle step-by-step reasoning mode”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 52

Use /fast

Use concise, fast response mode. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /fast produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /fast, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /fast. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/fast returns a visible report, state change, export/import result, or controlled action matching “Use concise, fast response mode”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 53

Use /elevated

Use expert-level detailed mode. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /elevated produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /elevated, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /elevated. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/elevated returns a visible report, state change, export/import result, or controlled action matching “Use expert-level detailed mode”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 54

Use /focus

Suppress unrelated work and focus on a task. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /focus produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /focus, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /focus. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/focus returns a visible report, state change, export/import result, or controlled action matching “Suppress unrelated work and focus on a task”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 55

Use /unfocus

Exit focus mode. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /unfocus produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /unfocus, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /unfocus. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/unfocus returns a visible report, state change, export/import result, or controlled action matching “Exit focus mode”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 56

Use /queue

Switch queue mode or queue a background task. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /queue produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /queue only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /queue status. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/queue returns a visible report, state change, export/import result, or controlled action matching “Switch queue mode or queue a background task”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 57

Use /side

Ask a side question without interrupting the main task. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /side produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /side, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /side What does this term mean?. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/side returns a visible report, state change, export/import result, or controlled action matching “Ask a side question without interrupting the main task”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 58

Use /btw

Alias for /side. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /btw produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /btw, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /btw What does this term mean?. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/btw returns a visible report, state change, export/import result, or controlled action matching “Alias for /side”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 59

Use /id

Alias for /whoami. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /id produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /id, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /id. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/id returns a visible report, state change, export/import result, or controlled action matching “Alias for /whoami”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 60

Use /whoami

Show current identity, channel, and session information. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /whoami produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /whoami, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /whoami. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/whoami returns a visible report, state change, export/import result, or controlled action matching “Show current identity, channel, and session information”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 61

Use /allowlist

Show or edit DM and channel allowlists. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /allowlist produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /allowlist, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /allowlist. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/allowlist returns a visible report, state change, export/import result, or controlled action matching “Show or edit DM and channel allowlists”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 62

Use /usage

Show current runtime usage and limits. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /usage produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /usage, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /usage. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/usage returns a visible report, state change, export/import result, or controlled action matching “Show current runtime usage and limits”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 63

Use /activation

Show or edit channel activation policy. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /activation produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /activation, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /activation. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/activation returns a visible report, state change, export/import result, or controlled action matching “Show or edit channel activation policy”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 64

Use /send

Send a message to a channel target. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /send produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /send only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /send <test-target> Ai Keeper connection test. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/send returns a visible report, state change, export/import result, or controlled action matching “Send a message to a channel target”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 65

Use /restart

Restart the active runtime or connector. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /restart produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /restart only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /restart status. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/restart returns a visible report, state change, export/import result, or controlled action matching “Restart the active runtime or connector”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 66

Use /vc

Show or manage voice-channel status. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /vc produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /vc, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /vc status. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/vc returns a visible report, state change, export/import result, or controlled action matching “Show or manage voice-channel status”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 67

Use /approve

Approve or deny pending exec or tool approvals. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /approve produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /approve only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /approve list. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/approve returns a visible report, state change, export/import result, or controlled action matching “Approve or deny pending exec or tool approvals”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 68

Use /subagents

List, spawn, or manage sub-agents. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /subagents produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /subagents only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /subagents list. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/subagents returns a visible report, state change, export/import result, or controlled action matching “List, spawn, or manage sub-agents”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 69

Use /kill

Terminate a running agent session or background task. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /kill produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /kill only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /kill <test-task-id>. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/kill returns a visible report, state change, export/import result, or controlled action matching “Terminate a running agent session or background task”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 70

Use /steer

Redirect conversation to a different agent or mode. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /steer produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /steer only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /steer <agent-name>. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/steer returns a visible report, state change, export/import result, or controlled action matching “Redirect conversation to a different agent or mode”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 71

Use /acp

Show or configure Agent Communication Protocol policies. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /acp produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /acp, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /acp status. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/acp returns a visible report, state change, export/import result, or controlled action matching “Show or configure Agent Communication Protocol policies”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 72

Use /export-session

Export a session transcript as Markdown or JSON. Learn the safe first invocation and the evidence that confirms it worked.

Beginner2–5 minResult · /export-session produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • No important unsaved generation should be in progress.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open Chat, type /export-session, and inspect the autocomplete description before pressing Return.

  2. Use the smallest safe example: /export-session markdown. Replace placeholders only with disposable test values.

  3. Read the command report in the transcript and compare it with the current Chat, model, session, or system state.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/export-session returns a visible report, state change, export/import result, or controlled action matching “Export a session transcript as Markdown or JSON”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.
Workflow 73

Use /import-session

Import a conversation from an exported JSON file. Learn the safe first invocation and the evidence that confirms it worked.

Intermediate2–5 minResult · /import-session produces its intended Chat result✓ SlashCommandHandler name and description verified
Before you start
  • A Chat conversation with a Ready model/provider route.
  • Use a disposable conversation, path, task, or destination for the first test.
  • Type /commands when you need the live syntax shown by the installed app.

  1. Open a disposable Chat and type /import-session only when you are ready to read its usage/confirmation without applying an important change.

  2. Use the smallest safe example: /import-session. Replace placeholders only with disposable test values.

  3. Read the preview, target, or approval prompt. Cancel if the target is broader than intended; otherwise confirm once and inspect the result.

  4. If the command returns usage text, follow that live syntax instead of guessing. The installed app is authoritative.

You are done when
/import-session returns a visible report, state change, export/import result, or controlled action matching “Import a conversation from an exported JSON file”; errors are reported in the transcript rather than silently ignored.
If something looks wrong
Unknown commandType /commands and choose the autocomplete entry; custom commands may use similar names.
The command needs argumentsUse the usage line returned by Chat and fill one placeholder at a time.
The target is important or too broadCancel and repeat with a disposable session, explicit ID, narrow path, or test destination.

Keep going

← Models, engines, and runtimesPrevious collectionAPIs, providers, and external clients →Next collection