2.3 KiB
2.3 KiB
SKILL: Policy Enforcer
Overview
The Policy Enforcer is the deterministic gate that ensures all probabilistic proposals adhere to the Core Invariants defined in the System Policy.
Architectural Intent: The Implicit Bridge
Unlike traditional software where a "Kernel" might have hardcoded rules, the Org-Agent harness is a "dumb" pipeline. This skill creates the bridge between human-readable rules and machine-enforced constraints through three mechanisms:
- Topological Bootstrapping: By declaring a #+DEPENDS_ON: dependency on the Policy file's ID, we ensure the System Policy is always registered in the Lisp image's skill catalog before this enforcer attempts to guard it.
- Priority Preemption: By setting :priority 1000, this skill registers itself as the very first check in the decide-gate. It effectively "pre-empts" all other skills, ensuring that no action (like a shell command or a file write) is even considered until it has cleared the alignment check.
- Decoupled Enforcement: The harness does not "know" it is enforcing a policy. It simply executes the highest-priority deterministic functions provided by its skills. This allows the Sovereign to swap out policies or enforcers without ever touching the core harness code.
Implementation
(defskill :skill-policy-enforcer
:priority 1000 ; Absolute highest priority
:trigger (lambda (context) t) ; Always active as a fallback
:probabilistic (lambda (context)
\"You are the Org-Agent Policy Enforcer. Your goal is to ensure all actions empower the user through the Lisp Machine and adhere to the System Policy.\")
:deterministic (lambda (action context)
;; Basic invariant check: Block actions that appear to violate sovereignty
(let ((payload (getf action :payload)))
(if (and payload (search \"proprietary\" (format nil \"~s\" payload)))
(progn
(org-agent:harness-log \"DETERMINISTIC [Policy]: Sovereignty violation suspected. Blocking action.\")
nil)
action))))