yolop

name	yolop
description	Built-in guide to using yolop, slash commands, keyboard shortcuts, CLI flags, tools, and session controls. Use when the user asks how yolop works, what commands or shortcuts are available, what yolop can do, or wants help using the terminal UI.
user-invocable	true

Yolop user guide

This skill is the durable reference for how to use yolop interactively. The live command list always comes from the session registry (/help in the TUI, or list_commands for the agent); this document explains what those commands do and how the UI behaves.

Quick reference in the TUI

• /help, commands with descriptions plus keyboard shortcuts
• /tools, comma-separated list of tools available this session
• /mcp, configured MCP servers (or where to add them)
• /cwd, workspace root
• Startup banner, workspace, model, tools, command names, and the most common shortcuts

Slash commands

Commands are contributed by capabilities. The interactive TUI registers all of the following; --print and ACP omit the terminal-only client commands.

Terminal client commands

These act on the TUI itself (clear transcript, open overlays, quit). The agent can also run them via run_command when the user asks in plain language.

Command	Description
/help	Show commands, descriptions, and keyboard shortcuts
/tools	List available tools for this session
/mcp	List configured MCP servers
/cwd	Print the workspace root
/status [compact|expanded|toggle]	Change session status bar layout (clicking the status bar toggles too)
/model [id]	Open the model list; an argument switches straight to a listed model
/effort [level]	Open the reasoning-effort picker; an argument pre-fills selection
/clear	Clear the transcript and re-show the startup banner
/shell <command>	Run a shell command from the workspace root with bounded inline output
/quit or /exit	Exit yolop

Shell aliases: !<command> and !shell <command> run the same bounded shell path as /shell (handy for quick one-offs).

Session and agent commands

Command	Description
/setup [subcommand]	Guided provider/model setup, or direct forms: status, provider, model, effort, token, url, attribution, approval
/goal [condition]	Set a completion condition and keep working until met; pause/resume, clear, or omit for status
/background	Show the task tree and branch usage
/btw <question>	Ask a side question with session context, no tools, not added to history
/worktree [off]	Show worktree status, or off to disable auto worktree activation

User-invocable skills (for example /yolop, /yolop-config, /skill-management) also appear in the registry when installed.

The model list

/model (and clicking the model in the status bar) opens your model list: an ordered, cross-provider menu of the models you switch between, spanning providers so one selection changes provider and model together. A "Browse all models…" row falls through to the old per-provider catalog for anything not on the list.

Entries whose provider you have not signed in to are shown and marked; picking one runs that provider's sign-in first, then switches to the model you picked.

Edit the list from a shell (it applies to the running session):

yolop models list
yolop models add openrouter anthropic/claude-opus-4-8 --label "Opus (OpenRouter)"
yolop models move claude-opus-4-8 1
yolop models rm gpt-5.4-mini
yolop models use gpt-5.6-sol
yolop models reset          # back to the built-in default list

It is stored as [[models]] in settings.toml, so it can also be edited by hand. Unset means the built-in default list, led by gpt-5.6-sol.

Keyboard shortcuts

Composer (idle)

Shortcut	Action
Enter	Send message
Shift-Enter	Insert newline in the composer
Alt-Shift-Enter	Send (alternative submit)
Tab	Accept the first slash-command suggestion, or insert a literal tab
← / →	Move cursor in the composer
/ then type	Command palette suggestions (accept with Tab)
Ctrl+V	Paste a clipboard image as an attachment, or attach large text as a placeholder
Terminal paste	Large pastes (>1000 chars) attach as [Pasted Content N chars] placeholders

Session control

Shortcut	Action
Ctrl-C (once, empty input)	Arm exit, press again to quit
Ctrl-C (with draft text)	Clear the composer
Ctrl-D	Exit immediately
Enter while a turn runs	Queue a steering message for the next turn boundary
Esc twice (while a turn runs)	Cancel the current turn
Ctrl+B	Toggle the interactive task-tree panel

Task-tree panel

Shortcut	Action
↑ / ↓ or k / j	Select a task
x or Delete	Request cooperative cancellation
Esc or q	Close the panel

Transcript history scrolls through the terminal's own scrollback (ratatui inline viewport); there is no in-app page-up/page-down.

Setup overlays (/setup, /model, /effort)

Shortcut	Action
↑ / ↓ or k / j	Move selection
Enter	Confirm
Esc	Cancel / go back
0–9	Jump to numbered entry (where shown)

/model opens the model list; its last row browses a provider's full catalog.

What yolop can do

yolop is an autonomous coding agent for the workspace it was started in.

• Files: read, write, edit, grep, delete, and map the repository
• Shell: bash -lc from the workspace root (120 s timeout, capped output)
• AST search: structural ast_grep across common languages (default harness)
• AST edit (opt-in), previewed ast_edit rewrites; enable with [[capabilities]] ref = "ast_edit"
• Background work: detached shell commands with completion wakes
• Web: web_fetch, free_web_search, and duckduckgo_instant_answer
• Memory: durable cross-session notes via remember / recall / forget
• Skills: SKILL.md packages in workspace, global, and bundled system scopes; find/install from skills.sh via search_skills / install_skill
• MCP: extra tools from .mcp.json / global mcp.json
• Hooks: block, allow, or audit tool calls (see yolop-hooks skill)
• Goal loops: /goal keeps working until an evaluator confirms the condition
• Soft approval: optional spoken consent before destructive steps (/setup approval)
• Sessions: resume with --session <id>; every run logs to events.jsonl

Natural-language control: the agent can switch model, effort, or provider, clear the screen, show help, and more without the user typing slash commands, see knowledge/specs/conversational-control.md.

CLI flags (non-interactive)

Flag	Description
-C, --cwd <PATH>	Workspace root (default: current directory)
--provider <P>	Force a provider
-m, --model <ID>	Override model for the chosen provider
-p, --print <PROMPT>	One-shot mode, print result and exit
--acp	Agent Client Protocol over stdio (editors like Zed)
--session <ID>	Resume a previous session
--session-dir <PATH>	Override where session folders are stored
--reasoning-effort <E>	Reasoning effort when the model profile supports it

Subcommands: yolop version, yolop into zed, yolop worktree list|prune.

Related skills

Skill	When to use
yolop-config	Change settings.toml, provider, tokens, models, capabilities
yolop-hooks	Configure behavioral hooks
skill-management	Install, upgrade, or remove skills
ast-grep	Structural code search and rewrite patterns

Answering help requests

When the user asks what yolop can do or how to use it:

1. Summarize the relevant section from this skill (commands, shortcuts, or features, match what they asked).
2. For the live command or tool list, prefer /help, /tools, or run_command / list_skills rather than guessing.
3. Point power users at README.md and knowledge/specs/ for protocol and configuration depth; keep conversational answers short.