@otto-code/protocol 0.8.7 → 0.8.9

package/dist/agent-personalities.js

@@ -112,14 +112,14 @@ export function summarizePersonalityForSelection(personality) {
     };
 }
 // The conductor's standing directive - the distilled `/epic` method taught to
-// the sole orchestrator role at spawn, so orchestration is emergent (the agent
-// recognizes team-shaped work and runs it) rather than something a user must
-// invoke. Kept here as one exported constant so the wording is testable and
-// shared. See projects/agent-orchestration/agent-orchestration.md.
-export const ORCHESTRATOR_METHOD_DIRECTIVE = "You are the orchestrator - the team's sole conductor. Team-shaped work is yours to run, and you should reach for it naturally, not only when asked. " +
-    "First apply the complexity gate: if a task is small and not splittable, just do it - no ceremony. Only orchestrate when the work is large, parallelizable, or benefits from independent perspectives. " +
-    "When you do orchestrate: (1) if the shape is unclear, dispatch a researcher to survey and a planner to draft a typed plan; (2) declare that plan as a Run with start_run - phases typed research/plan/implement/design/verify/gate/deliver, fanning out candidates where several angles help and attaching a judger to grade them, looping until enough pass; (3) put a gate before irreversible or costly steps so the user approves; (4) synthesize the passing results into the deliverable. " +
-    "Prefer start_run over hand-spawning and tracking agents yourself - the runtime fans out, gathers typed verdicts, and enforces the loop for you. Every phase maps to a teammate's role; if the active team lacks a role a phase needs, say so plainly and stop rather than papering over the gap.";
+// the sole orchestrator role at spawn. It chooses direct work, a dedicated
+// worker, a follow-up task, or a Run by the task's actual needs rather than
+// treating orchestration as the default. Kept here as one exported constant so
+// the wording is testable and shared. See projects/agent-orchestration/agent-orchestration.md.
+export const ORCHESTRATOR_METHOD_DIRECTIVE = "You are the orchestrator - the team's sole conductor. Choose tools because the task needs their specific capability, never because a tool is available or named. " +
+    "Do a small, self-contained task directly. Use create_agent only for an independently executable piece of active work that benefits from a dedicated worker. Use spawn_task only to preserve a concrete, out-of-scope follow-up for later. Use start_run only when the active work needs a declared multi-agent plan with daemon-managed fan-out, gathering, judging, loops, or approval gates. " +
+    "For work that genuinely needs orchestration: (1) if the shape is unclear, dispatch a researcher to survey and a planner to draft a typed plan; (2) declare that plan as a Run with start_run - phases typed research/plan/implement/design/verify/gate/deliver, fanning out candidates where several angles help and attaching a judger to grade them, looping until enough pass; (3) put a gate before irreversible or costly steps so the user approves; (4) synthesize the passing results into the deliverable. " +
+    "Every phase maps to a teammate's role; if the active team lacks a role a phase needs, say so plainly and stop rather than papering over the gap.";
 /**
  * The in-context "role directive" injected into a personality's system prompt at
  * spawn. The orchestrator gets the full conductor method; other coordinators
@@ -137,7 +137,7 @@ export function composeRoleFocusDirective(roles) {
         return `${ORCHESTRATOR_METHOD_DIRECTIVE} (Your roles: ${roleList}.)`;
     }
     if (normalized.some((role) => PERSONALITY_ROLE_INFO[role].tier === "coordinator")) {
-        return `You are a coordinator personality (roles: ${roleList}). You front interactive work and may delegate: use list_personalities to see who else is available, and spawn other agents or hand off to the team's orchestrator whenever delegating gets the work done faster or better.`;
+        return `You are a coordinator personality (roles: ${roleList}). You front interactive work and may delegate when the task benefits from it: use list_personalities to see who else is available, then either do the work directly, spawn a dedicated worker for an independent active piece, or hand off genuinely multi-agent work to the team's orchestrator.`;
     }
     return `You are a focused worker personality (roles: ${roleList}). Someone is waiting on this specific task - stay on it and finish it. You can still call list_personalities to see the roster, but don't spawn sub-agents or start side workflows unless it is genuinely essential to completing this job.`;
 }

package/dist/agent-types.d.ts

@@ -79,6 +79,8 @@ export interface AgentModelDefinition {
     id: string;
     label: string;
     description?: string;
+    /** Catalog-owned model-family identity, currently emitted by Otto Brain only. */
+    family?: string;
     isDefault?: boolean;
     metadata?: AgentMetadata;
     contextWindowMaxTokens?: number;