Skip to content

Latest commit

 

History

History
308 lines (239 loc) · 13.1 KB

File metadata and controls

308 lines (239 loc) · 13.1 KB

Slash commands

Built-in commands typed in the input box. The descriptions in this table are the canonical source of truth and match the slash-suggestion popup and the /help output exactly.

Project and user-defined commands are covered under Custom commands.

Built-in commands

Command Description
/models Switch the active model
/connections Manage LLM provider connections
/mcp Manage MCP servers (enable/disable, reconnect)
/compact Compact older complete rounds now
/new Start a new session, keeping the current one in history. Typing the retired /clear (or /reset) suggests /new instead — it never wipes anything in place
/permissions [clear] Show or clear always-allowed tool rules
/autopilot [on|off] Toggle autopilot mode (agent runs without human intervention)
/principal <code|architect|reviewer|security> Switch the principal role — changes persona and capability scope
/review Run an on-demand session-review diagnostic of the current round
/search <query> Semantic search over the project's session history
/session [status|list|resume|fork|open|new] Manage durable sessions
/sessions Browse past sessions
/dashboard Open the session dashboard — a full-screen live view over every daemon session (console + sessions dock), with preview / attach / interrupt / prompt / create (ADR-0096; layout per ADR-0097). /host is a hidden alias
/btw Open a side conversation that runs alongside the main session
/resume [id] Resume the most recent or selected session
/repeat [cron prompt|list|cancel id] Schedule a prompt on a cron expression (cron-only alias for /schedule)
/schedule [when prompt|list|cancel id] Schedule a prompt: cron (recurring) or countdown/absolute-time (one-shot)
/init [path] Initialize a .neenee/ config tree
/reload Re-read config.toml and apply changes live (MCP servers, permissions, bash policy, hooks)
/trust Trust this project's .neenee/config.toml (MCP servers + hooks) and load them
/untrust Revoke trust for this project (disconnects MCP, unloads hooks)
/skills [list|reload] List or reload available skills
/skill <name> Load a skill by name
/tools Toggle individual session tools on or off
/config Open user configuration
/export Export the current conversation to the clipboard as Markdown
/debug trace [on|off] Toggle per-project provider round-trip tracing for debugging
/debug preview Dry-run the next request body to a file (no provider call)
/help Show available commands and keybindings
/exit Exit the program

Several interactive management commands, including /models, /connections, /permissions, /tools, /mcp, /skills, and /config, are handled in the TUI. Commands that mutate agent or session state are dispatched to the backend.

Trigger-word suggestions ("did you mean …")

Retired commands and common synonyms are not executable — there is no /clear, /reset, or /continue. Typing one pins a suggestion row on top of the completion popup pointing at the supported command, and accepting the row rewrites the input to it:

You type Suggested Why
/clear /new Clearing the transcript in place was removed; a fresh session keeps the old one on disk
/reset /new Same fresh-session semantics
/continue /resume Picks a session up where it left off

The mapping is presentation-only and lives in one table (TRIGGER_WORD_SUGGESTIONS in neenee-transport/src/startup.rs); adding a row extends the steering without growing the executable command surface.

/serve

Superseded by the unified daemon (ADR-0096). The standalone neenee serve / neenee-server daemon now owns every session and serves them all over the control plane; hot-attaching a listener to a single running TUI session is a legacy of the per-session-host model. Use neenee serve to run the daemon and neenee attach / /dashboard to drive its sessions. See the Server WebSocket API for the current protocol.

Subcommands

/schedule

Form Effect
/schedule <when> <prompt> Schedule <prompt> to run at <when> (see below)
/schedule list List scheduled jobs (id, kind, trigger, next fire, prompt)
/schedule cancel <id> Cancel a scheduled job
/schedule help Show syntax help

<when> is one of:

  • a cron — five fields minute hour day month weekday, recurring (e.g. */5 * * * * every 5 min, 0 9 * * 1-5 09:00 on weekdays);
  • a countdown — one or more <number><unit> pairs from now (10m, 2h30m, 1d12h, in 10 minutes, in 2 hours 30 minutes; units: s/m/h/d and their long forms);
  • an absolute timeHH:MM today (or tomorrow if already passed), today HH:MM, tomorrow HH:MM, tomorrow, at HH:MM, YYYY-MM-DD HH:MM, or YYYY-MM-DDTHH:MM.

Cron jobs recur (and fire their first run immediately); countdown and absolute jobs fire once and are then removed. Jobs are durable (survive restarts). Recurring cron jobs auto-expire after 30 days. /schedule is the clock-driven scheduler for autopilot, reminders, and one-shot timers.

/repeat

Form Effect
/repeat <cron> <prompt> Schedule <prompt> on the five-field <cron> and run it now (cron-only alias for /schedule)
/repeat list List scheduled jobs (id, kind, trigger, next fire, prompt)
/repeat cancel <id> Cancel a scheduled job
/repeat help Show cron syntax help

/repeat is retained as a cron-only alias for /schedule. Use /schedule for countdown (10m) or absolute-time (14:00, tomorrow 09:00) one-shots.

/session

Form Effect
/session status Show session id, parent, counts, checkpoint, compaction
/session list List durable session branches
/session resume [id] Resume the most recent or selected session
/session fork Fork the current conversation into a child session
/session open <id-prefix> Open a session by id or id prefix
/session new Start a new durable session

/permissions

Form Effect
/permissions List always-allowed tool rules for this process
/permissions clear Clear process-local always-allow rules

/tools

Form Effect
/tools Open the tools manager overlay

Opens a centered, scrollable list of every tool available to the live session — builtins and mcp:<server> tools — each with its source and an [on]/[off] badge. / move the selection, Space toggles a tool on or off (the harness applies it and replies with a fresh snapshot), and Esc closes. /tools is handled entirely in the TUI and is never forwarded to the backend.

/config

Form Effect
/config Open the Settings overlay

The Settings overlay exposes Appearance and Layout. Appearance offers the zen, midnight, nord, catppuccin, and paper presets. The Custom option opens an eight-field #RRGGBB editor for background, surface, text, muted, accent, success, warning, and error colors. Valid custom colors preview live; Enter saves and applies the palette, while Esc cancels the draft. Changes apply immediately and persist in the [tui] table of config.toml.

/autopilot

Form Effect
/autopilot Toggle autopilot on/off
/autopilot on Run without human intervention (no confirmations, no questions)
/autopilot off Restore interactive prompts

When on, the agent acts without human intervention: tool permissions auto-approve before write/execute tools (bash, write_file, edit_file, …), the ask_user question tool is reclaimed (hidden from the model; any stale call short-circuits), interactive command stdin is closed instead of prompting, and the system prompt is told no human is reachable. Affects the live process only. For the design intent and every surface the flag enforces, see Autopilot operation.

/principal

Form Effect
/principal <role> Switch the active principal role (persona + capability scope)
/principal List the available roles and the current one

Switches the live agent's principal role at runtime (ADR-0053). Each role is a preset over the product's base identity — the mission/persona shifts, the product identity stays. It can also be triggered mid-message with the @principal:<role> mention:

Role Scope
code The default coding principal — full capabilities, unrestricted writes
architect Design and review focus — full read, writes retained but the persona steers toward analysis and written rationale before changes
reviewer Read-only code review — read/search/inspect tools only (no write_file, edit_file, or bash)
security Read-only, command-confined security audit — read/search plus a narrow command allowlist

Unknown role names are rejected with the list of valid roles.

/btw

Form Effect
/btw Open a side conversation that runs alongside the main session

Opens a lightweight side conversation for asking quick questions without disturbing the main session context (ADR-0017).

/review

On-demand only — triggers a bounded read-only REVIEW envoy that diagnoses the current round and reports verdicts. /review takes no arguments. The original periodic-cadence design (ADR-0016) was superseded by on-demand review plus in-loop steering (ADR-0030); ADR-0016 itself remains Accepted.

/search

Form Effect
/search <query> Semantic search over the project's session history

Returns the most relevant past messages for the query (see the search_history tool). Useful for recalling earlier decisions.

/skills

Form Effect
/skills List available skills (alias for /skills list)
/skills list List available skills
/skills reload Rescan local skill directories and refetch remote repositories

/skill

Form Effect
/skill <name> Load a skill by name into the conversation context

/init

Form Effect
/init [path] Initialize a .neenee/ config tree; path defaults to .

/reload

Form Effect
/reload Re-read config.toml and apply changes live

Re-loads configuration without restarting: MCP servers, permissions, bash_policy, and hooks are all applied to the live process.

/trust and /untrust

Form Effect
/trust Trust this project's .neenee/config.toml (MCP servers + hooks) and load them
/untrust Revoke trust for this project (disconnects MCP, unloads hooks)

A project's .neenee/config.toml is only loaded after you trust the project; /trust grants that once, and /untrust revokes it. The trust posture is defined in ADR-0085.

/export

Form Effect
/export Render the live conversation as Markdown — metadata header (session id, provider/model, exported-at), then a chronological transcript of user prompts, assistant replies, tool calls, and inlined tool results — and copy it to the system clipboard so it can be pasted into another agent to continue the work.

The receiving agent gets the full chain of decisions and side effects: hidden and system messages are skipped (mirroring TUI rendering), reasoning traces are folded into collapsible <details> blocks, and envoy transcripts nested under envoy results are summarised by message counts instead of dumped in full. If the system clipboard is unavailable, the export falls back to OSC52 or surfaces the underlying clipboard error.

Custom commands

Markdown files discovered in .neenee/commands/ (project-local, higher priority) and $XDG_DATA_HOME/neenee/commands/ (user-global, XDG; default ~/.local/share/neenee/commands/). The legacy pre-XDG fallback ~/.neenee/commands/ was removed (ADR-0013 → ADR-0058); the filename stem or frontmatter name becomes /name after lowercasing and stripping a leading /. Names allow ASCII letters, digits, -, and _.

See Paths for the full override stack and the project-vs-XDG boundary.

Optional YAML frontmatter:

---
name: review
description: Review changes
---

The template body supports $ARGUMENTS (the full argument string) and $1 through $9 positional placeholders. Built-in command names are reserved and are not shadowed by custom commands.

See also

  • Harness architecture — the round loop, durable session, permission broker, context compaction
  • Modals — the /models, /connections, and /sessions pickers