▣ wi

Push llmmsg protocol mechanics into MCP send tool + hub (shrink model-held context)

Ref

#616 (#616)

Project

llmmsg-srv

Status

backlog

Priority

normal

Type

feature

Assigned

coder-llmmsgsrv-cc coder

Created by

—

Created

2026-06-03T04:55:25.227Z

Updated

2026-06-15T08:39:09.291Z

Sub-items

Questions

Event log

• 2026-06-03T04:55:25.227Z · created
  wi cli
• 2026-06-03T04:55:39.201Z · scope
  Goal: models hold less llmmsg-srv protocol context by moving MECHANICS server-side (judgment stays in model). Scope: 1. Hub: auto-inherit origin_aro from re=<tag> - on a reply, if origin_aro unset, set it from the replied-to message's origin_aro. Kills the empty-origin_aro drop/flood class (this week's bug). 2. MCP send tool: encode defaults in the schema/description - DM is default; full-ARO requires explicit to=aro:X (already) but make kind an enum with one-line hints; don't require kind for DMs. 3. Hub: SOFT-nudge (not block) on mis-targeted sends - e.g. kind=dm to aro:X, or a human-DM sent from inside an ARO without fanout. Nudge only; hub guessing intent + rejecting is worse than a flood. 4. Trim docs/llmmsg-protocol.md prose once the tool encodes the defaults (keep judgment guidance, drop mechanics). Consensus: PM + coder-chatduo-cc. Elazar approved 2026-06-03. Design first, PM sequences vs other WIs before build.
• 2026-06-15T08:39:04.409Z · assigned · wi-cli-whey
  coder-llmmsgsrv-cc / coder
• 2026-06-15T08:39:09.291Z · assigned · wi-cli-whey
  coder-llmmsgsrv-cc / coder