memex/notes/v1-0-skill-audit.org

#+TITLE: Sovereign OS v1.0 Core Skill Audit Report
#+DATE: [2026-04-09 Thu]
#+AUTHOR: Gemini CLI Agent
#+FILETAGS: :audit:v1-0:sovereign-os:psf:

* Executive Summary
This audit evaluated all 39 Org files in the =projects/org-agent/skills/= directory against the five core criteria for Sovereign OS v1.0. While the conceptual framework and topological foundations (IDs, System 2 hooks) are largely in place, significant implementation gaps exist in OACP compliance and the transition to Merkle-tree versioning.

** Overall Statistics
- *Total Skills Scanned:* 39
- *Topological Integrity:* ~85% (Most files have unique IDs; some depend on name-based strings)
- *System 2 Safety:* ~70% (Hooks exist, but symbolic verification logic is often stubbed)
- *Persistence:* ~20% (Basic object-store interaction exists, but Merkle-tree logic is MISSING)
- *OACP Compliance:* ~5% (Only 2/39 skills use =def-cognitive-tool=)
- *Sovereignty Mandate:* ~95% (High adherence to local-first, Lisp-native primitives)

* Critical Skill Deep Dive

** org-skill-agent (The Kernel)
- *Status:* STABLE / MANDATORY
- *Topological Integrity:* PASS (ID: 47425a43-2be0-423c-8509-22592cfe9c9e)
- *GAPS:*
  - Missing =def-cognitive-tool= for kernel introspection.
  - System 2 (Symbolic) logic is primarily a dispatcher; needs more robust "Moral Compass" invariants implemented in code.

** org-skill-safety-harness (The Gatekeeper)
- *Status:* FUNCTIONAL / MANDATORY
- *Topological Integrity:* PASS (ID: 98576df2-c496-4e4a-9acb-0bca514a0305)
- *GAPS:*
  - Whitelist is comprehensive but needs a mechanism for skills to dynamically (and safely) register their own symbols.
  - Missing =def-cognitive-tool= for security telemetry.

** org-skill-self-fix (The Repair Mechanism)
- *Status:* PARTIAL / EXTENSION
- *Topological Integrity:* PARTIAL (Uses =skill-scientist= string instead of ID in =#+DEPENDS_ON:=)
- *GAPS:*
  - Rollback logic depends on "Interactive Steering" snapshots which are not yet fully integrated with the Merkle-tree architecture.
  - Missing =def-cognitive-tool=.

** org-skill-object-store-persistence (The Memory Image)
- *Status:* STUBBED / MANDATORY
- *GAPS:*
  - *CRITICAL:* Currently uses simple =maphash= + =print= serialization. Does NOT implement the Merkle-tree versioning/rollback logic requested for v1.0.
  - Missing background sync triggers beyond basic heartbeat.

** org-skill-ast-normalization (The Structure Guardian)
- *Status:* BLUEPRINT-ONLY / MANDATORY
- *GAPS:*
  - *CRITICAL:* Implementation is missing. Functions like =ast-normalize-file= and =find-conflicts= are defined in the blueprint but not tangled/implemented.
  - No integration with the Object Store for structural change tracking.

* Categorized Skill Audit

** Core Mandatory
Foundational skills required for system boot and basic operation.
| Skill | ID | Gaps |
|-------+----+------|
| org-skill-agent | 47425a43... | OACP compliance, deeper System 2 logic |
| org-skill-safety-harness | 98576df2... | Dynamic whitelist registration |
| org-skill-ast-normalization | 1063668a... | Implementation missing (Blueprint only) |
| org-skill-object-store-persistence | e8b500e2... | Missing Merkle-tree/Rollback logic |
| org-skill-lisp-machine-bootstrap | 81e3492e... | Purely research/simulation stage |
| org-skill-auth-api-key | ab7f8ca4... | OACP compliance |
| org-skill-router | 0fb9e9cb... | Complexity classification is basic |
| org-skill-sub-agent-manager | e870d860... | Needs better thread isolation logic |

** Provider
Interfaces to LLM backends.
| Skill | ID | Gaps |
|-------+----+------|
| org-skill-provider-ollama | f605bf22... | OACP compliance, model discovery |
| org-skill-provider-gemini | 52799ee8... | OACP compliance |
| org-skill-provider-openai | c6cbd603... | OACP compliance |
| org-skill-provider-anthropic | a44d29c6... | OACP compliance |

** Actuator
Sovereign interfaces for system modification.
| Skill | ID | Gaps |
|-------+----+------|
| org-skill-shell-actuator | 0ae190ec... | Missing fine-grained permissions |
| org-skill-org-mode | f7db1884... | Needs better AST-to-JSON bridge integration |
| org-skill-inbound-gateway | 89ee87e0... | Multi-channel normalization is incomplete |

** Extension
Feature-rich skills extending the kernel.
| Skill | ID | Gaps |
|-------+----+------|
| org-skill-latent-reflection | latent-reflection | (OACP PASS) Needs more triggers |
| org-skill-memory-archivist | 98923a43... | (OACP PASS) IPFS restore is manual |
| org-skill-self-fix | 65891ce2... | Implementation/Rollback logic |
| org-skill-model-explorer | ae49bb72... | Implementation missing (Overview only) |

* Global Strategic Gaps

** 1. Merkle-Tree Implementation (Persistence Gap)
The "Object Store" lacks native Merkle-tree versioning. Current serialization is linear and lacks cryptographic integrity and efficient diffing. This is a priority for Component III.

** 2. OACP Compliance (Tooling Gap)
95% of skills do not use =def-cognitive-tool=. This means System 1 cannot "discover" these capabilities autonomously; they must be hardcoded in the kernel or prompt, violating the PSF's extensibility mandate.

** 3. Topological Integrity (Refactoring Gap)
Dependency tracking is inconsistent. Some skills use =id:UUID=, others use =skill-name=. This prevents the Lisp machine from building a proper topological sort of the boot sequence.

** 4. Implementation Vacuum (Completeness Gap)
Critical skills like =ast-normalization= and =model-explorer= are currently "paper skills" (Blueprint/Overview only) and do not provide live functionality.

* Recommendations
1. *Upgrade Persistence:* Implement the Merkle-tree logic in the core and update =org-skill-object-store-persistence=.
2. *Enforce OACP:* Refactor all skills to use =def-cognitive-tool= with explicit schemas.
3. *Normalize Dependencies:* Standardize all =#+DEPENDS_ON:= to use UUIDs.
4. *Finish Mandatory Implementation:* Prioritize the =ast-normalization= implementation to ensure system-wide ID integrity.