Stage 2: Reason (reason.lisp)

Stage 2: Reason (reason.lisp)

Stage 2: Reason (reason.lisp)

Architectural Intent: Unified Cognition

The Reason stage is the cognitive engine of the Org-Agent. It unifies two distinct reasoning modes:

Probabilistic Reasoning: Consulting neural models (LLMs) to generate action proposals based on current context.
Deterministic Reasoning: Running those proposals through a series of deterministic safety gates (Policy, Invariants, and Skill-specific validation) to ensure alignment and security.

Package and Registry

We initialize the probabilistic backends and the provider cascade which determines the order in which models are consulted.

(in-package :org-agent)

(defvar *probabilistic-backends* (make-hash-table :test 'equal))
(defvar *provider-cascade* nil)
(defvar *model-selector-fn* nil)
(defvar *consensus-enabled-p* nil)

(defun register-probabilistic-backend (name fn)
  "Registers a neural provider (e.g., :gemini, :anthropic) with its calling function."
  (setf (gethash name *probabilistic-backends*) fn))

Neural Dispatch (Probabilistic)

The `probabilistic-call` function manages the cascade of neural providers. If the primary provider fails, it automatically falls back to the next one in the list.

(defun probabilistic-call (prompt &key (system-prompt "You are the Probabilistic engine.") (cascade nil) (context nil))
  "Dispatches a neural request through the provider cascade. Returns a Lisp plist or a failure log."
  (let ((backends (or cascade *provider-cascade*)))
    (or (dolist (backend backends)
          (let ((backend-fn (gethash backend *probabilistic-backends*)))
            (when backend-fn
              (harness-log "PROBABILISTIC: Attempting backend ~a..." backend)
              (let* ((model (when *model-selector-fn* (funcall *model-selector-fn* backend context)))
                     (result (if model 
                                 (funcall backend-fn prompt system-prompt :model model)
                                 (funcall backend-fn prompt system-prompt))))
                ;; If the result is valid and doesn't contain a failure log, return it.
                (unless (or (null result) 
                            (and (stringp result) (search ":LOG" result)))
                  (return result))))))
        ;; Final fallback if all backends in the cascade fail.
        (list :type :LOG :payload (list :text "Neural Cascade Failure: All providers exhausted.")))))

Cognitive Proposal (Think)

The `think` function represents the "intuitive" side of the agent. It identifies the active skill, assembles the global context, and asks the probabilistic engine for a structured action proposal.

(defun think (context)
  "Generates a Lisp action proposal based on current context."
  (let* ((active-skill (find-triggered-skill context))
         (tool-belt (generate-tool-belt-prompt))
         (global-context (context-assemble-global-awareness))
         (assistant-name (or (uiop:getenv "MEMEX_ASSISTANT") "Agent")))
    (if active-skill
        (let* ((prompt-generator (skill-probabilistic-prompt active-skill))
               (raw-prompt (when prompt-generator (funcall prompt-generator context)))
               (system-prompt (format nil "IDENTITY: Actuator for ~a. MANDATE: ONE Lisp plist. ~a ~a" 
                                      assistant-name global-context tool-belt)))
          (if (and raw-prompt (> (length raw-prompt) 1))
              (let* ((thought (probabilistic-call raw-prompt :system-prompt system-prompt :context context))
                     ;; Ensure we are working with a string for read-from-string
                     (cleaned (if (stringp thought) (string-trim '(#\Space #\Newline #\Tab) thought) thought)))
                (if (stringp cleaned)
                    (handler-case (read-from-string cleaned)
                      (error (c) (list :type :EVENT :payload (list :sensor :syntax-error :code cleaned :error (format nil "~a" c)))))
                    cleaned))
              (list :type :LOG :payload (list :text (format nil "Skill '~a' triggered (Deterministic only)" (skill-name active-skill))))))
        nil)))

Deterministic Verification

Once a proposal is generated, it MUST pass through the deterministic gates. Every skill can register a gate that inspects and potentially modifies or blocks the proposed action.

(defun deterministic-verify (proposed-action context)
  "Iterates through all skill deterministic-gates sorted by priority."
  (let ((current-action proposed-action)
        (skills nil))
    ;; 1. Collect and sort active gates
    (maphash (lambda (name skill) (declare (ignore name)) (when (skill-deterministic-fn skill) (push skill skills))) *skills-registry*)
    (setf skills (sort skills #'> :key #'skill-priority))
    
    ;; 2. Execute gates sequentially
    (dolist (skill skills)
      (let ((gate (skill-deterministic-fn skill)))
        (setf current-action (funcall gate current-action context))
        ;; If any gate returns a LOG or EVENT, it has intercepted the action.
        (when (and (listp current-action) (member (getf current-action :type) '(:LOG :EVENT)))
          (harness-log "DETERMINISTIC: Intercepted by skill '~a'" (skill-name skill))
          (return-from deterministic-verify current-action))))
    current-action))

The Reason Gate

The entry point for the Reason stage. It orchestrates the transition from probabilistic "thought" to deterministic "verification."

(defun reason-gate (signal)
  "Unified Stage: Combines Probabilistic proposals and Deterministic verification."
  ;; Only process events that haven't been reasoned yet.
  (unless (eq (getf signal :type) :EVENT) (return-from reason-gate signal))
  
  (let ((candidate (think signal)))
    (if candidate
        (setf (getf signal :approved-action) (deterministic-verify candidate signal))
        (setf (getf signal :approved-action) nil))
    (setf (getf signal :status) :reasoned)
    signal))

6.1 KiB Raw Blame History

Stage 2: Reason (reason.lisp)