Dirigent Plugin

Hook strategy

message:received caches a per-session decision from deterministic rules.
before_model_resolve applies providerOverride + modelOverride when decision says no-reply.
before_prompt_build prepends end-marker instruction + scheduling identifier instruction when decision allows speaking.

Rules (in order)

non-discord -> skip
bypass sender -> skip
end symbol matched -> skip
else -> no-reply override

Config

See docs/CONFIG.example.json.

Required:

noReplyProvider
noReplyModel

Optional:

enabled (default true)
discordOnly (default true)
listMode (human-list | agent-list, default human-list)
humanList (default [])
agentList (default [])
channelPoliciesFile (per-channel overrides in a standalone JSON file)
schedulingIdentifier (default ➡️) — moderator handoff identifier
enableDirigentPolicyTool (default true)
multiMessageStartMarker (default ↗️)
multiMessageEndMarker (default ↙️)
multiMessagePromptMarker (default ⤵️)

Unified optional tool:

dirigent_tools
- Discord actions: channel-private-create, channel-private-update, member-list
- Policy actions: policy-get, policy-set-channel, policy-delete-channel
- Turn actions: turn-status, turn-advance, turn-reset
bypassUserIds (deprecated alias of humanList)
endSymbols (default ["🔚"])
enableDiscordControlTool (default true)
Discord control actions are executed in-plugin via Discord REST API (no discordControlApiBaseUrl needed)
discordControlApiToken
discordControlCallerId
enableDebugLogs (default false)
debugLogChannelIds (default [], empty = all channels when debug enabled)

Per-channel policy file example: docs/channel-policies.example.json.

Policy file behavior:

loaded once on startup into memory
runtime decisions read memory state only
direct file edits do NOT affect memory state
dirigent_tools policy actions update memory first, then persist to file (atomic write)

Moderator handoff format

When the current speaker NO_REPLYs, the moderator bot sends: <@NEXT_USER_ID>➡️

This is a non-semantic scheduling message. The scheduling identifier (➡️ by default) carries no meaning — it simply signals the next agent to check chat history and decide whether to speak.

Multi-message mode / shuffle mode

Human sends the configured start marker (default ↗️) → channel enters multi-message mode.
While active, agents are forced to no-reply and the moderator sends only the configured prompt marker (default ⤵️) after each additional human message.
Human sends the configured end marker (default ↙️) → channel exits multi-message mode and normal scheduling resumes.
No separate moderator "entered/exited mode" confirmation message is sent; the markers themselves are the protocol.
The first moderator message after exit uses the normal scheduling handoff format: <@NEXT_USER_ID>➡️.
/dirigent turn-shuffling, /dirigent turn-shuffling on, and /dirigent turn-shuffling off control per-channel reshuffling between completed rounds.

Slash command (Discord)

/dirigent status
/dirigent turn-status
/dirigent turn-advance
/dirigent turn-reset
/dirigent turn-shuffling
/dirigent turn-shuffling on
/dirigent turn-shuffling off

Debug logging:

set enableDebugLogs: true to emit detailed hook diagnostics
optionally set debugLogChannelIds to only log selected channel IDs

3.5 KiB Raw Blame History