PSF: Foundry Progress Sync. 57 high-fidelity blueprints established. Open Fleet routing (Kimi/Qwen) active. GTD updated.

notes/org-skill-economist.org

@@ -5,7 +5,7 @@
 #+DEPENDS_ON: skill-router skill-performance-auditor
 
 * Overview
-The *Economist Agent* manages the PSF's compute resources. It predicts the "Cost of Thought" for a task and autonomously selects the most cost-effective LLM provider or local model based on complexity and remaining budget.
+The *Economist Agent* manages compute resources by prioritizing high-performance open-source models (Kimi, Qwen) to minimize costs while maintaining architectural integrity.
 
 * Phase A: Demand (PRD)
 :PROPERTIES:
@@ -13,12 +13,7 @@ The *Economist Agent* manages the PSF's compute resources. It predicts the "Cost
 :END:
 
 ** 1. Purpose
-Optimize token usage and compute overhead without sacrificing architectural integrity.
-
-** 2. User Needs
-- *Predictive Budgeting:* Estimate token cost before triggering SOTA models.
-- *Provider Switching:* Dynamically route tasks between local (Ollama) and cloud (Gemini) models.
-- *Audit Reports:* Provide transparency on compute consumption.
+Optimize token usage by leveraging open-weights models via OpenRouter.
 
 * Phase B: Blueprint (PROTOCOL)
 :PROPERTIES:
@@ -26,38 +21,27 @@ Optimize token usage and compute overhead without sacrificing architectural inte
 :END:
 
 ** 1. Architectural Intent
-The *Economist Agent* provides cost-governance for the Neural Engine. It intercepts `think` requests and determines the optimal Model/Provider based on task complexity, priority, and current budget constraints.
+Dynamically route tasks to the "Open Fleet" (Kimi/Qwen).
 
 ** 2. Semantic Interfaces
 
-*** Routing Logic (2026 Fleet)
+*** Routing Logic (2026 Open Fleet)
 #+begin_src lisp :tangle ../projects/org-skill-economist/src/economist-logic.lisp
 (in-package :org-agent)
 
 (defun economist-route-task (context)
-  "Analyzes the stimulus context and returns a prioritized list of providers.
-   High-priority or complex tasks (e.g., :architect) get powerful models.
-   Routine tasks (e.g., :heartbeat, :persistence) get cheap/flash models."
-  (let* ((payload (getf context :payload))
-         (sensor (getf payload :sensor))
-         (complexity (ignore-errors (uiop:symbol-call :org-agent.skills.org-skill-router :router-classify-complexity context))))
-    (cond
-      ;; Explicit user interaction or Reasoning tasks
-      ((or (member sensor '(:user-command)) (eq complexity :REASONING)) '(:openrouter))
-      
-      ;; Cognitive or Reflexive tasks
-      (t '(:openrouter))))) ; Route through OpenRouter to avoid direct Google 429s
+  (declare (ignore context))
+  '(:openrouter)) ; Exclusively use OpenRouter for the Open Fleet
 
 (defun economist-get-model-for-provider (provider &optional context)
-  "Returns the specific model ID recommended for the given provider/complexity.
-   Updated for April 2026 SOTA. Prefers Gemini 3.0/2.5 Flash for reflexes."
+  "Returns Open-Source SOTA model IDs. Updated April 2026."
   (let ((complexity (ignore-errors (uiop:symbol-call :org-agent.skills.org-skill-router :router-classify-complexity context))))
     (case provider
       (:openrouter
        (case complexity
-         (:REASONING "anthropic/claude-3.5-sonnet")
-         (:COGNITION "moonshotai/kimi-k2.5")
-         (t "google/gemini-3-flash-preview")))
+         (:REASONING "qwen/qwen-2.5-72b-instruct") ; Heavy lifting
+         (:COGNITION "moonshotai/kimi-k2.5")      ; Interaction
+         (t "qwen/qwen-2.5-72b-instruct")))       ; Standard reflex
       (t nil))))
 #+end_src
 
@@ -72,9 +56,10 @@ The *Economist Agent* provides cost-governance for the Neural Engine. It interce
 
 * Registration
 #+begin_src lisp
-(defskill :skill-economist
-  :priority 100 ; High priority to ensure cost-checks happen first
-  :trigger (lambda (context) (eq (getf (getf context :payload) :sensor) :cost-audit))
-  :neuro (lambda (context) nil)
-  :symbolic (lambda (action context) (economist-route-task context)))
+(progn
+  (defskill :skill-economist
+    :priority 100
+    :trigger (lambda (context) (eq (getf (getf context :payload) :sensor) :cost-audit))
+    :neuro (lambda (context) nil)
+    :symbolic (lambda (action context) (economist-route-task context))))
 #+end_src