Kubrick v0.14.0: Deterministic Contract Consolidation

Kubrick v0.14.0 Release Notes

Deterministic Contract Consolidation

Kubrick v0.14 turns the existing symbolic-cinematic compiler into a more dependable
standalone Hermes skill. This release emphasizes reproducibility, dependency clarity,
structured failure, provider invariants, installation safety, and official Hermes
packaging rather than adding new symbolic ontology.

Highlights

One canonical manifest

kubrick.manifest.yaml now governs runtime profiles, intents, actions, aliases,
recipes, artifact and schema registries, providers, authority classes, and exit codes.
The CLI router derives its public registries from this file.

Explicit runtime profiles

• stdlib: manifest, routing, skill validation, diagnostics, and release checks
• validation: YAML and JSON Schema validation plus compile/adaptation integration
• dev: full regression, governance, repeatability, installer, and corpus tooling

Kubrick remains a Hermes skill loaded from its directory. pyproject.toml documents
tooling contracts and does not make package installation mandatory.

Reproducible receipts

Compile receipts now include Kubrick, corpus, schema-bundle, provider-adapter,
normalized-input, and command identities. Mutation controls prove that each digest
changes when its governed input changes, while canonical normalized outputs remain
byte-identical across repeated runs.

Structured failure

The router and compiler emit schema-valid diagnostics with stable exit codes:

• 2: invalid command
• 3: required optional dependency unavailable
• 4: NOT_COMPUTABLE

Human-readable stderr remains the default. Set KUBRICK_DIAGNOSTICS=json for machine
consumers.

Provider semantic preservation

Generic, Grok Imagine, Flux, SD3, and Midjourney adapters emit preservation reports
covering graph identity, required content, ownership, geometry, state change, residue,
continuity, and negative constraints. Critical semantic loss fails closed as
PROVIDER_LOSS.

Atomic Hermes installation

The Bash installer now supports staged validation, atomic activation, external
backups, install receipts, --dry-run, --rollback, and --version. A failed staged
validation or activation leaves the prior installation intact.

Modern Hermes skill contract

SKILL.md now uses the required Hermes section order and a concise activation-focused
description. Official packaging includes the storyboard recipe fixtures and a
repository-level contract test.

Verification summary

The release gates cover:

• Linux, macOS, and Windows
• Python 3.10 through 3.14
• stdlib and validation profiles
• 22 Hermes behavioral and adversarial evaluations
• manifest/router parity and stable diagnostics
• deterministic receipt identity and repeatability
• provider semantic preservation and loss injection
• installer fresh install, upgrade, rollback, and failure safety
• outcome governance, Wave 2/3, design-specification, and authority boundaries
• canonical and exported Hermes skill layouts

Compatibility

• Existing do <intent> commands and legacy aliases remain supported.
• Continuity Forge, MCP, image providers, and vision providers remain optional.
• OpenClaw remains on its permanent host-specific branch and requires a follow-up
  contract-alignment pass tracked by issue #32.

Release documents

• docs/ROADMAP-v0.14.md
• docs/RELEASE-CHECKLIST-v0.14.md
• CHANGELOG.md

v0.13.0 — Forge Feedback · Multi-Provider Adapters · Operator Surface

Kubrick 0.13.0

Wave 2 and Wave 3 of the production-hardening roadmap are shipped.

Tracking: Issues #3 and #4 closed via PR #24. Docs aligned in PR #25.

Highlights

• Forge multi-signal feedback — ledger diffs, revisions, saturation, collisions, ingestion, and payoff outcomes → deterministic observation bundles (forge-signals)
• Multi-signal evolution — confidence, mutation success, feasibility, anti-slop, cultural boundaries, payoff → proposal-only evolution with human review gates
• First-class project ledgers — pattern history, rehydrate/apply-forge, retrieval snapshots; Forge remains canonical
• Multi-provider adapters — Grok Imagine, Flux, SD3, Midjourney share one latent graph; syntax-only translation
• Closed-loop visual QA — geometry / state / residue / convergence reported separately
• CLI + optional MCP operators — saturation, counterpoint, convergence lock, surface-occult audit, motif mutation, symbolic-architecture export
• Design specification compiler — design-build for schema-valid design specs without silent authority promotion
• Time-sensitive cultural-signal packs — contemporary memetic patterns with provenance

Install

git clone https://github.com/scrimshawlife-ctrl/Kubrick.git
cd Kubrick
git checkout v0.13.0
./install.sh

Default skill path: ~/.hermes/skills/kubrick

Verify

python scripts/kubrick.py validate-skill
python scripts/kubrick.py validate-corpus
python scripts/test_wave2_wave3.py
python scripts/audit_release_version.py --strict

Docs

• README
• QUICKSTART
• Release notes
• Roadmap
• Changelog

Compatibility

• External model APIs, vision APIs, MCP, and Continuity Forge remain optional
• Evolution proposals never apply automatically
• Adapters do not rewrite canonical symbolic intent or graph identity

v0.10.0 — Neuro-Symbolic & Compositional Upgrades

Kubrick v0.10.0

Neuro-Symbolic & Compositional Upgrades

This release brings structured, scene-graph-inspired symbolic engineering to the Kubrick skill while keeping all esoteric and symbolic elements fully latent and observable-only.

Highlights

• Internal Motif/Structure Graph Layer (neuro-symbolic conditioning)
• Explicit Disentanglement of Layout/Geometry, Semantics/Function, Attributes/States
• Compositional Layered Encoding (CMA/MLS analogs)
• Conditioning-Style Prompt Engineering
• Neuro-Symbolic Predicate Validation
• Enhanced Single-Frame State Modeling
• Full Esoteric-Alchemical Encoding Lexicon (200+ entries)
• Updated Dynamic Selection Engine and docs

See CHANGELOG.md for complete details.