Chapter 10 — Slash commands

Slash commands are the control plane. Type / followed by a name to run a command instead of sending the line to the model. Type /help any time to see the full list.

CLI and GUI are peers. Every command in this chapter works identically from the CLI REPL, the GUI’s Terminal tab, and the GUI’s Chat tab — the /<word> input goes through the same dispatcher in all three surfaces. A few commands that mutate tool state (/mcp add, /skill install, /plugin install, /kms use) even activate their effects in the current session without a restart; the table notes which ones.

Resolution order

When you type /<word>, thClaws resolves it in this order:

1. Built-in command — the table below.
2. Installed skill — rewrites the line into a Skill(name: "word") invocation (Chapter 12).
3. Legacy prompt command — .md template from a commands/ directory, with $ARGUMENTS substituted (this chapter).
4. Unknown — yellow error.

First match wins. Skills never shadow built-ins because built-ins are tried first.

Built-in command reference

Session & model

Command	What it does
/help	Show all built-in commands
/model [NAME]	Show current model, or switch to NAME (validated; typos revert)
/models	List available models from the current provider — prints fully-qualified routable ids (e.g. openrouter/google/gemma-3-27b-it:free) so any row pastes straight into /model <id>. Non-chat models (audio/image generation) are filtered out automatically; OpenRouter rows also filter when “Free only” is on in Settings. /models refresh re-seeds from the upstream provider list
/provider [NAME]	Show current provider, or switch
/providers	List every provider + its default model
/save	Force-save the current session to disk
/load ID\|NAME	Load a session by id, id-prefix, or title
/sessions	List saved sessions (newest first)
/rename [NAME]	Rename the current session (no arg clears the title)
/fork	Save the current session, then start a fresh one seeded with a summary of it — a clean context that remembers where you were
/resume ID\|NAME	(CLI flag --resume) restart with a session loaded
/clear	Wipe in-memory history (doesn’t touch saved files)
/history	Print a message-count summary
/translate [--language=<code>] <text or file>	Translate text or a file in a side conversation — an alias for /agent translator …, so it never touches your session
/summarize [--language=<code>] <text>	Summarise in a side conversation — an alias for /agent summarizer …. --language sets the summary’s language, not the source’s
/extract <url or file>	Clip a page or file into clean markdown, images downloaded alongside. Runs isolated, so the raw page never enters your context — the point of it for long pages
/compact	Summarise history to free tokens
/cwd	Show the working directory (sandbox root)
/pwd	Same thing — the shell name for it

Memory & context

Command	What it does
/memory	List memory entries
/memory read NAME	Print a memory entry
/context	Show the combined system prompt (project + agents + skills catalog)

Tools, skills, plugins, MCP

Command	What it does
/marketplace [--refresh]	Open the unified marketplace browser (GUI modal: skills · MCP · plugins · subagents). CLI prints a combined summary
/skills	List loaded skills
/skill show NAME	Full description + path for a skill
/skill marketplace [--refresh]	Browse the catalog at thclaws.ai/api/marketplace.json
/skill search QUERY	Substring-search the marketplace catalog
/skill info NAME	Marketplace detail for one skill (license, source, install URL)
/skill install [--user] <name-or-url> [name]	Install a skill — bare slug → marketplace lookup, otherwise git or .zip URL
/mcp marketplace [--refresh]	Browse hosted + installable MCP servers in the catalog
/mcp search QUERY	Substring-search the MCP marketplace
/mcp info NAME	MCP marketplace detail (transport, command/url, license)
/mcp install [--user] NAME	Install a marketplace MCP — clones source if needed, writes mcp.json entry
/plugin marketplace [--refresh]	Browse the plugin catalog
/plugin search QUERY	Substring-search the plugin marketplace
/plugin info NAME	Marketplace detail for one plugin (use /plugin show NAME for installed)
/subagent marketplace [--refresh]	Browse subagents (agent defs) in the catalog
/subagent search QUERY	Substring-search the subagent marketplace
/subagent info NAME	Marketplace detail for one subagent
/subagent install [--user] <name-or-url> [name]	Install a subagent .md — bare slug → marketplace lookup, otherwise git or .zip URL
/<skill-name> [args]	Invoke an installed skill directly
/<command-name> [args]	Invoke a legacy prompt command (template)
/plugins	List installed plugins (enabled + disabled)
/plugin install [--user] <url>	Install a plugin bundle
/plugin remove [--user] <name>	Uninstall a plugin
/plugin enable [--user] <name>	Enable a disabled plugin
/plugin disable [--user] <name>	Disable without uninstalling
/plugin show <name>	Manifest details
/mcp	List active MCP servers and their tools
/mcp add [--user] <name> <url>	Register a remote (HTTP) MCP server
/mcp remove [--user] <name>	Remove an MCP server from config

Knowledge bases (KMS)

Command	What it does
/kms (or /kms list)	List every discoverable KMS; * marks ones attached to this project
/kms new [--project] NAME	Create a new KMS (default scope is user)
/kms use NAME	Attach a KMS to this project’s chats
/kms off NAME	Detach a KMS
/kms entry [NAME] [--set SLUG \| --clear]	Show or set the page the KMS browser opens on. Recorded when the KMS’s first page is created; inferred when nothing is recorded
/kms verify [NAME] [--llm] [--fix]	Evidence check: citations resolve, archived sources still carry their quotes, nothing asserts a number uncited. --llm adds a per-page entailment audit; --fix unwraps links written inside URLs
/kms rename OLD NEW	Rename a KMS folder; the attachment follows (also: right-click a KMS in the sidebar)
/kms drop NAME [--force]	Delete a KMS; dry-run without --force (also: sidebar right-click → Delete…)
/kms show NAME	Print the KMS’s index.md
/kms maintain NAME [--apply]	One-command maintenance: structural fixes + source reconciliation vs live sessions + stale refresh + contradiction reconciliation, in one staged pass. Dry-run by default. GUI-only. Alias: tidy
/kms html NAME [OUT]	Generate a single-file interactive HTML site from a KMS (v0.8.5+). Agent reads the KMS via tools, designs components, writes <OUT>/index.html (default ./<NAME>-site/) in your workspace
/kms export-okf NAME [OUT]	Export a KMS as an Open Knowledge Format (OKF) bundle to ./NAME-okf/ (or OUT) for portable interchange
/kms import-okf BUNDLE NAME [--project]	Create a new KMS from an OKF bundle folder (default scope is user)
/dream [FOCUS]	Consolidate the project’s KMS by mining recent sessions (GUI-only, dispatches a built-in side-channel agent)

See Chapter 9 for the full KMS concept + workflow, including the /kms html HTML export, OKF import/export (also on the sidebar’s right-click “Knowledge” menu), graph view, and the /dream consolidation flow.

Background research

Command	What it does
/research <query>	Spawn a background research job — multi-iteration web search + multi-page KMS write
/research [--kms NAME] [--max-notes N] [--max-iter K] [--novelty 0.X] [--worker-model ID] [--append] [--dry-run] [--budget-time T] <query>	Start with overrides (see ch20)
/research (or /research list)	List all jobs (newest first)
/research status ID	Detailed view (phase, iteration, score)
/research show ID	Print synthesized result in chat
/research cancel ID	Cancel a running job; partial result discarded
/research wait ID	Block CLI prompt until terminal (CLI-only)
/policy (or /policy status)	Active org policy: source file, issuer, expiry, which blocks are on, and audit sinks with their dropped-record counts (Enterprise)

See Chapter 20 for the full pipeline + KMS layout + flag reference.

Agent behaviour

Command	What it does
/permissions MODE	Switch between auto and ask mid-session
/thinking 0-3	Thinking level for every provider: 0 off · 1 low · 2 medium · 3 high · auto (provider default). Also /thinking <tokens> for a raw budget. Same knob as the think pills under the model chip; persists to thinkingBudget
/tasks	List tasks / todos the agent has created
/plan [enter\|exit\|status]	Toggle plan mode — read-only exploration, then step-gated execution. status reports the current plan without changing mode (Chapter 18)
/config key=val	Override a config value for this session only
/agent NAME PROMPT	Spawn a user-driven side-channel subagent (GUI-only, runs concurrently with main)
/agents	List active background side-channel agents (id, name, elapsed)
/agent cancel ID	Cancel a running side-channel agent
/agent new NAME	Open the GUI editor to author a new agent def (.thclaws/agents/NAME.md) — GUI-only
/agent edit NAME	Open the GUI editor for an existing/built-in agent def — GUI-only
/dream [FOCUS]	Dispatch the built-in dream agent to consolidate KMS (GUI-only) — see Chapter 9
/team	Attach to the team tmux session (or show team status)
/doctor	Run diagnostic checks
/cost	Session spend, by provider and model
/usage	Token usage by provider and model
/version	Show the thClaws version and commit SHA
/system [stats \| grep <pattern>]	Print the system prompt actually being sent this turn. stats shows its size breakdown by section; grep <pattern> searches it — useful for “is my AGENTS.md actually in there?”
/reload	Re-read settings, skills, agents and MCP config without restarting
/reload-prompt	Re-read only the prompt templates
/quit	Exit (aliases: /exit, /q). In the GUI, opens a native confirm dialog (“Quit?”) before closing — Cancel keeps the session open

Automation

Each of these has its own chapter; the table is the quick reference.

Command	What it does
/schedule list \| show <id> \| run <id> \| pause\|resume <id> \| rm <id>	Manage recurring (cron / interval / watch) jobs — see Chapter 19
/loop <interval> <body> | status | stop	Repeat a message on a fixed interval (default 5 min). One loop at a time — see Chapter 31
/goal start <objective> [--budget-tokens N] [--budget-time T] [--auto] [--require <path>] | status | show | continue | complete | abandon	A long-running objective the agent audits itself against, with budgets and hard limits. Pairs with /loop, or drives itself with --auto — see Chapter 31
/workflow run <goal> \| exec <path> \| list \| inspect <id> \| resume <id> \| rm <id>	Author, review and run a multi-agent workflow script — see Chapter 25

Cloud and deployment

Command	What it does
/cloud list [--mine] \| get <slug> \| status	Browse and install agents from the thClaws.cloud catalog — see Chapter 27
/publish <file.html>	Put a self-contained HTML page on the web at a private URL that expires in 3 days. Needs a thClaws.cloud token and a credit balance above zero
/deploy [--pod URL] [--token T] [--dry-run] [--full] [--no-restart]	Ship this workspace’s .thclaws/ to a remote pod. --dry-run first

Learning

Command	What it does
/quiz <topic\|url\|file>	Generate a study quiz from a topic, a URL or a local file, then play it

Shell escape

Command	What it does
! <command>	Run <command> in the terminal directly, bypassing the agent

Useful for quick sanity checks (! ls, ! git status) without spending model tokens.

Skill and command shortcuts

Any installed skill is callable as /<skill-name>:

❯ /skills
  docx — Create, read, edit Word documents
  pdf  — Read, split, merge, OCR PDFs
  …

❯ /pdf extract text from report.pdf
(/pdf → Skill(name: "pdf"))
Using the pdf skill to extract text from report.pdf…

Legacy prompt commands live as markdown files:

# .thclaws/commands/review.md
---
description: Code review a branch
---
Review the diff from `main` to HEAD. Flag security issues, bad naming,
and missing tests. Focus on $ARGUMENTS.

❯ /review authentication
(/review → prompt from .thclaws/commands/review.md)
Reviewing the diff, focused on authentication…

$ARGUMENTS expands to whatever came after the command name. If the template has no placeholder and the user typed args, they’re appended on a blank line.

Writing your own slash commands

For quick one-liners, drop an .md file into .thclaws/commands/. For anything with scripts or scaffolding, make it a skill (Chapter 12). For a whole bundle (skills + commands + MCP), ship it as a plugin (Chapter 16).