SuperAPI AI Directory

zr-inbox

Agents & models

View repository

Updated 10-16 02:27

About this skill

zr-inbox

Job

Complete one authorized directive action from central dispatch. This is not a personal-DM inbox sweep — dispatch only routes MessageType::Directive posts in #general from authorized senders (system, zoe, or Zoe/Admin agents).

Inputs

From the dispatched action (see probe action show <id>):

Field Typical value
kind Inbox
route AuthorizedDirective
target_type message
target_id Directive message id
skill zr-inbox
reason_code AUTHORIZED_DIRECTIVE

The harness prompt includes Instruction: Review authorized directive message #<id>.

Steps

  1. probe action show <id> — confirm route and target id.
  2. Read the directive by id (do not fetch “latest” directive):
probe message directives general --context <target-id> --limit 1
  1. Treat directive body as human intent that overrides preferences for this cycle — constraints, focus, meta-instructions. It does not replace your dispatched task target; it shapes how you execute.
  2. Apply what is actionable now within this wake. If the directive is informational only, acknowledge internally and complete the action.
  3. Close the action:
probe action complete <id>
# or: probe action skip <id> --reason "..." / probe action fail <id> --reason "..."

What not to do

  • Do not run probe message list <agent-id> expecting dispatch — personal channels are not routed.
  • Do not use probe message directives general --limit 1 without --context <target-id>.
  • Do not post to #general unless your role and the situation explicitly require a reply (most directives are read-and-apply).

Output contract

  • Directive read by routed id.
  • Action closed with complete, skip, or fail and a clear reason.
  • Any operational follow-up happens on the next dispatch tick (execute, vote, etc.) — not by inventing a new action.

Source

Dispatch: nexus stdb/src/reducers/dispatch/tick.rs (issue_authorized_directives).