NEWv0.25.0 — Agent-built dashboards, multi-modal media, desktop app →

Chat Commands

Chat Commands

Chat commands are interactive commands sent directly in conversation across any supported interface (Matrix, Slack, Discord, Telegram, IRC, Nostr, TUI, and Web UI).

Invocation & Prefixes

Both slash (/) and bang (!) prefixes are supported interchangeably:

/status
!status
  • Runtime-handled: Intercepted at the message queue layer (enqueueMessage). They are never sent to the LLM, cost zero tokens, and execute immediately.
  • Unified output: Output is formatted cleanly (typically structured YAML or monospace text) and broadcast back to the calling interface.
  • Matrix / Slack / Discord friendly: The ! prefix prevents native client autocomplete from swallowing or hijacking commands (e.g., Matrix or Slack clients that intercept /status).

Core Commands

/status (or !status)

Displays the runtime status of the active agent:

  • Runtime version & agent name
  • Model, provider, and reasoning status
  • Process uptime and PID
  • Active session message count and raw file size
  • Token usage & cost breakdown
  • Active interface connections (Telegram, Slack, Matrix, Discord, IRC, Nostr)
  • Message queue state
!status

/wyd (or !wyd)

Surfaces what the agent is currently working on:

  • If idle, reports waiting for input.
  • If processing a turn, returns a concise summary of the active user request, current step progress, and recent tool invocations.
!wyd

/plugins (or !plugins)

Shows status, registered tools, routes, and active configuration for all loaded plugins (MCP, Skills, Dashboards, Sub-agents, Recall, Media, Slack, Matrix, Telegram, Discord, etc.).

!plugins

/restart (or !restart)

Restarts the agent process.

  • Replies Restarting. first, so the calling interface sees confirmation before shutdown.
  • Then spawns kern restart <agent directory> as a detached process, using the same node binary the agent is running on and this package's own entry point. The directory is passed, never the agent name, and PATH is never consulted, so it works from any daemon environment.
  • That process stops the agent via <dir>/.kern/agent.pid and starts it again; the TUI re-reads the port from <dir>/.kern/ and reconnects.
  • Registered as a Telegram bot command (appears in Telegram's / menu).
  • Agents cannot invoke this command themselves; it must be typed by a human operator.
!restart

/help (or !help)

Lists all available chat commands with descriptions, including builtin runtime commands and dynamically loaded plugin commands.

!help

Plugin Commands

/recall-health (or !recall-health)

Provided by the recall plugin.

Runs an instant, read-only diagnostic on the active session's vector embedding index in recall.db:

  • Indexed message scan progress and true vector coverage percentage
  • Chunk size distribution (percentiles and token buckets)
  • Oversized batch blockers (>8192 tokens or >16k chars)
  • UTF-16 surrogate integrity (isWellFormed)
  • Virtual table vector invariants (ghost vectors or missing rows)
  • Stalled tail identification (exact message index and role blocking the pipeline)
  • Overall 0–100 health score
!recall-health

/segment-health (or !segment-health)

Provided by the recall plugin.

Runs an instant, read-only diagnostic on the active session's semantic summary DAG in recall.db:

  • Segment tiling invariant per level (L0, L1, L2): overlaps, gaps, stragglers, and orphans
  • Injected summary waste percentage under the agent's real context budget
  • Parent/child consistency check (range mismatches, child outside parent)
  • Overall 0–100 health score
!segment-health

/skills (or !skills)

Provided by the skills plugin.

Lists all discovered skills across local workspace (skills/), installed directory (.agents/skills/), and bundled skills, showing their active/inactive status.

!skills

/subagents (or !subagents)

Provided by the subagents plugin.

Lists active and completed sub-agent tasks spawned during the current agent process lifetime, showing their ID, status, step count, execution time, and the interface/channel their result is delivered to.

!subagents

/mcp (or !mcp)

Provided by the mcp plugin.

Shows configured Model Context Protocol (MCP) servers, connection states, and registered external tools.

!mcp

HTTP API

GET /commands

Agents expose an authenticated endpoint returning all active chat commands (builtins + loaded plugins) as a JSON dictionary mapping command names to descriptions.

Used by the Web UI for dynamic autocomplete.

{
  "/status": "show agent status",
  "/restart": "restart agent daemon",
  "/help": "show available slash commands",
  "/plugins": "show status and tools for all plugins",
  "/skills": "list available skills and active status",
  "/recall-health": "show recall embedding health and pipeline stalls",
  "/segment-health": "show semantic segment summary tree health and rollup invariants",
  "/subagents": "list active and recent sub-agents",
  "/mcp": "show MCP server connection status and tools"
}