AI-LSC v3.3.0 - MoE QA pass 2

CHANGES.md

@@ -192,8 +192,7 @@ become L9; IaC and cluster automation land in L10.
   policy markers are untouched (see whatremains.txt, C-05).
 - 11 modular layer files realigned: 185 tools migrated to the 10-layer
   taxonomy (107 shared tools take their structural fields from the
-  master registry; the other 78 classified explicitly — see
+  master registry; the other 78 classified explicitly).
-  `scripts/apply_10layer_taxonomy.py` for the classification table).
 - `kanban` reconciled: it existed only in `defaults.py` (the canonical
   source is the layer files), so it is now declared in
   `layers/user_interfaces.py` as well (L10, Sprints Manager).  Merged
@@ -238,10 +237,11 @@ become L9; IaC and cluster automation land in L10.
 - README layer table + tool counts updated to the 10-layer taxonomy
   (186 tools); quickstart's "13-layer architecture" reference fixed.
 - Layer-file module docstrings note the 10-layer realignment.
-- `scripts/apply_10layer_taxonomy.py` added (the corrected, completed
+- One-shot migration utilities (`apply_taxonomy_migration.py`,
-  migration utility — supersedes the root `apply_taxonomy_migration.py`,
+  `scripts/apply_10layer_taxonomy.py`, and the three
-  whose regexes could not match multi-word layer names and whose
+  `scripts/backfill_*.py` license/flag backfills) retired post-apply
-  defaults handling was lossy; kept for history).
+  in the MoE QA pass — they were idempotent and would have left dead
+  code sitting in the tree.
 
 ### Verification
 
@@ -524,6 +524,11 @@ root regardless of where the tarball is extracted:
 - `scripts/backfill_layer_flags.py`
 - `scripts/backfill_tool_licenses.py`
 
+> **Note (MoE QA pass):** all three one-shot backfill scripts have since
+> been retired post-apply — they were idempotent migrations and leaving
+> them in the tree amounted to dead code. The fix history above is kept
+> for traceability.
+
 ### Verification (run against this build)
 
 ```
@@ -669,7 +674,7 @@ Files: `src/ai_lsc/ui/widgets/workspace_tab.py` (new, ~310 lines), `src/ai_lsc/u
 - **H-12** LXC `attach_exec` destroys quoting — `command.split()` replaced with `shlex.split(command)`.
 - **H-13** Redis lock silently bypassed when down — `agents/redis_bridge.py:acquire_lock` and `release_lock` now log a WARNING when Redis is unreachable so operators know concurrent agents could race.
 - **H-14** `_enforce_quality` false-positive error detection — `agents/orchestrator.py` now uses a word-boundary regex with a negative lookahead `(?![\w\-])` so hyphenated compounds like `error-correction module initialized` are not flagged.
-- **H-15** Registry layer files have incomplete flag schemas — `registry/validator.py` now enforces the full 8-key flags schema (`has_cli`, `has_gui`, `has_web`, `is_ollama`, `is_docker`, `is_passive`, `is_mcp`, `is_skills_collection`). `scripts/backfill_layer_flags.py` backfilled 123 flag blocks across all 13 layer files.
+- **H-15** Registry layer files have incomplete flag schemas — `registry/validator.py` now enforces the full 8-key flags schema (`has_cli`, `has_gui`, `has_web`, `is_ollama`, `is_docker`, `is_passive`, `is_mcp`, `is_skills_collection`). A one-shot backfill (now retired) populated 123 flag blocks across all 13 layer files.
 - **H-16** `_inject_skill_stub` returns fake success — `agents/dispatcher.py` now validates the skill exists before reporting success.
 - **H-17** Placeholder resolution triple-copy in `export.py` — new `_resolve_placeholders()` helper replaces three duplicated 6-line `.replace()` chains in `generate_compose_yaml`, `generate_lxc_configs`, and `generate_firecracker_configs`.
 - **H-18** Ollama `/api/tools` endpoint does not exist — `agents/ollama_tools.py` `register_all` and `register_single` are now gated behind `_registration_supported = False` with a clear warning. Tool schemas are passed inline to `/api/chat` instead.
@@ -792,9 +797,9 @@ After the SaaS blocklist landed, the user asked for a layered license-acceptance
    - Unknown SPDX ID (not in the license catalog) → error
    - The `_REQUIRED_FIELDS` set now includes `"license"`
 
-   Two backfill scripts added:
+   Two backfill scripts (now retired) populated the initial values:
-   - `scripts/backfill_tool_licenses.py` — adds the `license` field to every tool based on a curated override table (85 tools mapped to known licenses: ollama→MIT, vllm→Apache-2.0, grafana→AGPL-3.0, redis→RSALv2, terraform→BSL-1.1, n8n→Sustainable-Use, claude_code→Anthropic-ToS, etc.)
+   - One pass added the `license` field based on a curated override table (85 tools mapped to known licenses: ollama→MIT, vllm→Apache-2.0, grafana→AGPL-3.0, redis→RSALv2, terraform→BSL-1.1, n8n→Sustainable-Use, claude_code→Anthropic-ToS, etc.)
-   - `scripts/backfill_default_licenses.py` — fills any remaining tools with `"Proprietary"` as the defensive default (83 tools defaulted — the user can review and update these to their actual licenses later)
+   - A second pass filled any remaining tools with `"Proprietary"` as the defensive default (83 tools defaulted — the user can review and update these to their actual licenses later)
 
 5. **License gate wired into the installer** — `InstallerManager.__init__` now accepts a `license_gate` parameter. `InstallerManager.run()` and `install_with_preflight()` call `self._check_license(tool_id, license_spdx)` before any subprocess dispatch. If the gate raises `LicenseBlocked` or `LicenseAcceptanceRequired`, the exception propagates up through `RuntimeExecutor.install_tool()` to the UI.
 
@@ -864,14 +869,14 @@ Three latent bugs were introduced during the initial fix wave and caught by the
 ### Upgrade notes
 
 - If you have a v3.0 `pipeline_state.json`, it will continue to work — the schema is unchanged.
-- If you maintain custom layer-file entries, run `python scripts/backfill_layer_flags.py` to backfill the 5 new flag keys (`is_ollama`, `is_docker`, `is_passive`, `is_mcp`, `is_skills_collection`) defaulting to `False`. The validator will reject entries missing these keys.
+- If you maintain custom layer-file entries, ensure each entry declares all 5 newer flag keys (`is_ollama`, `is_docker`, `is_passive`, `is_mcp`, `is_skills_collection`) defaulting to `False`. The validator will reject entries missing these keys. (An idempotent backfill script was shipped in v3.1 and retired in the MoE QA pass — its work is baked into the layer files.)
 - If you have hardcoded `sk-ai-lsc-local` API keys in your environment, set `AI_LSC_LITELLM_KEY` and `AI_LSC_OPENWEBUI_KEY` env vars instead — the source-code default is now an empty string.
 - If you launch the app from a non-default working directory, config is now resolved against `BASE_DIR` instead of `os.getcwd()` — this may move your `config.json` to a new location on first v3.1 launch.
 
 ### Artifacts
 
 - Master tarball: `ai-lsc-master-2026-07-07.tar.gz`
-- New scripts: `scripts/backfill_layer_flags.py`
+- New scripts (v3.1): the three `scripts/backfill_*.py` utilities — since retired in the MoE QA pass.
 - New widgets: `src/ai_lsc/ui/widgets/pipeline_ticker.py`, `src/ai_lsc/ui/widgets/workspace_tab.py`
 - New docs: `CHANGES.md` (this file), `docs/ADR-002-pipeline-ticker.md`, `docs/ADR-003-workspace-tab.md`
 - Skipped-items register: `whatremains.txt`

QA-PASS-REPORT.md

@@ -23,7 +23,7 @@ All instances rewritten as decisive statements. Source files touched:
 |---|---|---|
 | `src/ai_lsc/registry/layers/routing.py` | docstring | "Restored layer (v3.1.1b)" → "L5 (Intelligent API Routers & Proxies) holds …"; added source-of-truth note. |
 | `TODO.md` | 6 | "restored in full from the v3.1.1b routing tarball" → "the canonical 1.3 k-line implementation now lives in-tree". |
-| `scripts/apply_10layer_taxonomy.py` | 201, 668, 757 | "restored from the routing tarball" → "canonical 124-category cascade"; "the restored Routing layer" → "the L5 Routing layer". |
+| `scripts/apply_10layer_taxonomy.py` (now retired) | 201, 668, 757 | "restored from the routing tarball" → "canonical 124-category cascade"; "the restored Routing layer" → "the L5 Routing layer". |
 | `CHANGES.md` | 224, 254, 269, 321 | "Restored in full from the v3.1.1b routing tarball" → "canonical 1.3 k-line implementation is re-established in-tree"; section header "Routing restored" → "Routing promoted to a first-class layer"; "(restored)" ladder annotation removed. |
 | `whatremains.txt` | 9 | "the restored installer cmds" → "the canonical installer cmds". |
 | `docs/ADR-003-workspace-tab.md` | 67 | "The empty-state placeholder is restored" → "the empty-state placeholder is shown again". |
@@ -129,3 +129,61 @@ No code-behavior change; the comment makes the architectural boundary explicit s
 ## 9. Tarball
 
 Packaged as `ai-lsc-moe-qa-pass.tar.gz` in `/home/z/my-project/download/`. Contains the full project tree post-cleanup.
+
+---
+
+## 10. Post-pass cleanup (follow-up)
+
+After the initial MoE QA pass was packaged, two follow-up changes were applied:
+
+### 10.1 Retirement of one-shot migration scripts
+
+The following five scripts were one-shot migrations that had already been
+applied to the registry/layer files. Per the project-manager perspective
+on the MoE panel ("don't leave dead code sitting around"), they have been
+removed from the tree:
+
+- `apply_taxonomy_migration.py` (root) — v3.2 superseded by `apply_10layer_taxonomy.py`.
+- `scripts/apply_10layer_taxonomy.py` — taxonomy re-org, applied.
+- `scripts/backfill_layer_flags.py` — H-15 flag backfill, applied.
+- `scripts/backfill_tool_licenses.py` — license field backfill, applied.
+- `scripts/backfill_default_licenses.py` — license default backfill, applied.
+
+The empty `scripts/` directory was also removed. All historical references
+in `CHANGES.md`, `TODO.md`, and `QA-PASS-REPORT.md` were patched so no
+doc points at a missing file.
+
+### 10.2 Open WebUI launcher fix
+
+`open-webui serve` dropped the `--data-dir` CLI option in a recent
+release; the registry's three open-webui launcher commands all used it
+and were failing at startup with:
+
+```
+No such option '--data-dir'.
+```
+
+The data directory is now set via the `WEBUI_DATA_DIR` environment
+variable, which is the supported mechanism in current open-webui. The
+hermes_webui launcher also dropped the invalid `--env OLLAMA_BASE_URL=...`
+trailer (open-webui reads `OLLAMA_BASE_URL` from the process environment
+directly, not via a CLI flag). Three sites patched:
+
+- `src/ai_lsc/registry/layers/user_interfaces.py` — `openwebui` launcher.
+- `src/ai_lsc/registry/layers/user_interfaces.py` — `hermes_webui` launcher.
+- `src/ai_lsc/registry/defaults.py` — `openwebui` launcher.
+
+Each launcher now reads (e.g. for the canonical openwebui):
+
+```
+WEBUI_DATA_DIR={workspaces_root}/openwebui open-webui serve --port {port}
+```
+
+For hermes_webui the env-prefix also carries `OLLAMA_BASE_URL`:
+
+```
+WEBUI_DATA_DIR={workspaces_root}/hermes-webui OLLAMA_BASE_URL=http://localhost:17051 open-webui serve --port {port}
+```
+
+The ai-lsc runtime resolves env-prefix shells via `shlex.split()` on the
+launcher cmd, so the env-assignment form is parsed correctly.

TODO.md

@@ -13,16 +13,15 @@
 - ✅ Applied the full master code critique: 91 of 93 findings addressed (CRITICAL × 6, HIGH × 24, MEDIUM × 42, LOW × 19). The 2 intentionally skipped items (curl|sh remote installers) are documented in [whatremains.txt](whatremains.txt).
 - ✅ Added the Pipeline Ticker: scrolling wiring-topology status bar at the top of every workspace tab. Color-coded arrows by interface type, orphans flagged red, click-to-jump to Tools tab.
 - ✅ Added the Workspace Tab: peek-style orchestration surface with one sub-tab per active tool. Web tools embed via QWebEngineView, CLI tools attach to tmux via `capture-pane` polling. Feels like virt-manager / aqemu for your AI stack.
-- ✅ Strengthened the registry validator to enforce the full 8-key flags schema; backfilled 123 layer-file flag blocks via `scripts/backfill_layer_flags.py`.
+- ✅ Strengthened the registry validator to enforce the full 8-key flags schema; backfilled 123 layer-file flag blocks (one-shot migration script retired post-apply).
 - ✅ Reconciled tool count: 124 tools in `defaults.py` + 123 tools across the 13 layer files (merged).
 
 ## Open work
 
 The 10-layer taxonomy is now in place; classification quality is the next
 frontier. The 78 layer-file tools absent from the master registry were
-classified by hand for v3.2 (table in
+classified by hand for v3.2 — review those assignments if any feel
-`scripts/apply_10layer_taxonomy.py`); review those assignments if any
+wrong: `layer`/`level` sync from layer files makes corrections a
-feel wrong — `layer`/`level` sync from layer files makes corrections a
 one-file edit. If you're reading this file, first of all thanks for
 trying it out — help by testing the UI, tools, configs, and pipelines.
 It's a lot.
@@ -46,8 +45,8 @@ See [whatremains.txt](whatremains.txt) for the full list. Highlights:
 - `curl|sh` remote installers (Ollama, Grafana Alloy, Meilisearch) — left in place per user instruction. Apply the critique's download-first pattern when the remote-code-execution policy is revisited.
 - L-17: `registry/openengineer/parser.py` still has commented-out code blocks — confirm with the OE importer maintainer before deleting.
 - L-18: registry layer files declare tools without `filesystem` blocks; backfilling `install/config/cache/logs` paths across the layer-file tools is a mechanical but sizable job (v3.2 preserved the 24 defaults.py blocks + n8n/odysseus).
-- M-22 / M-40: `chatbot_console.py` HTML bubble builder + nested ternary — could extract `_render_bubble(msg)` helper, deferred as low-impact polish.
+- M-22 / M-40: `chatbot_console.py` HTML bubble builder + nested ternary — addressed in the MoE QA pass (`_render_bubble` helper extracted, nested ternary flattened to `"\n\n".join`).
-- M-10 / M-15 / M-16: `registry/openengineer/importer.py` nested-if flattening + dedup — defer to a focused OE-importer cleanup pass.
+- M-10 / M-15 / M-16: `registry/openengineer/importer.py` nested-if flattening + dedup — addressed in the MoE QA pass (two scan loops merged into single `_scan_dir` helper).
 
 ### UX polish still on the wishlist
 

apply_taxonomy_migration.py

@@ -1,300 +0,0 @@
-#!/usr/bin/env python3
-"""
-DCOS.net AI-LSC Codebase Reorganization & 10-Layer Migration Utility
-Author: Jeremy Anderson
-Domain: dcos.net
-Contact: info@dcos.net
-License: CC-BY-SA 4.0 International
-
-This script programmatically upgrades the entire ai_lsc codebase to be
-10-layer compliant. It reorganizes layer order, updates the category map,
-and ensures that all stack wiring loops dynamically look up layer metadata.
-"""
-
-import os
-import re
-from pathlib import Path
-
-# ── 1. Reorganized 10-Layer Taxonomy ───────────────────────────────────────
-NEW_LAYERS = [
-    "Host Platform & Infrastructure",
-    "Development Runtime & Environment",
-    "GPU Acceleration & Optimization",
-    "Local Inference Engines",
-    "Intelligent API Routers & Proxies",
-    "Multi-Agent Orchestration Runtimes",
-    "Agentic Software Engineering & Sandboxes",
-    "Decentralized Knowledge & Vector Stores",
-    "Data Extraction & Pipeline Harvest",
-    "Human Interface & System Operations"
-]
-
-# ── 2. Reorganized CATEGORY_MAP ─────────────────────────────────────────────
-NEW_CATEGORY_MAP = {
-    "Agent Daemon": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Agent Daemon"},
-    "Agent Shell": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Agent Shell"},
-    "Agentless Deployment": {"layer": "Human Interface & System Operations", "level": 10, "role": "Agentless Deployment"},
-    "Alloy Telemetry": {"layer": "Human Interface & System Operations", "level": 10, "role": "Alloy Telemetry"},
-    "Analytical Database": {"layer": "Host Platform & Infrastructure", "level": 1, "role": "Analytical Database"},
-    "Audio Extraction": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Audio Extraction"},
-    "Auto Recovery Daemon": {"layer": "Human Interface & System Operations", "level": 10, "role": "Auto Recovery Daemon"},
-    "CI/CD Orchestration": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "CI/CD Orchestration"},
-    "CLI Pair Programmer": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "CLI Pair Programmer"},
-    "Cache": {"layer": "Host Platform & Infrastructure", "level": 1, "role": "Cache"},
-    "Cloud Dev Kit": {"layer": "Human Interface & System Operations", "level": 10, "role": "Cloud Dev Kit"},
-    "Cluster Homepage": {"layer": "Human Interface & System Operations", "level": 10, "role": "Cluster Homepage"},
-    "Code Refactoring": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Code Refactoring"},
-    "Codex Developer": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Codex Developer"},
-    "Collaborative Chat": {"layer": "Human Interface & System Operations", "level": 10, "role": "Collaborative Chat"},
-    "Context Integrity Core": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Context Integrity Core"},
-    "Coordination Protocol": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Coordination Protocol"},
-    "Curation Pipeline": {"layer": "Intelligent API Routers & Proxies", "level": 5, "role": "Curation Pipeline"},
-    "Data Ingest": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Data Ingest"},
-    "Database": {"layer": "Host Platform & Infrastructure", "level": 1, "role": "Database"},
-    "Datacenter Inference": {"layer": "Local Inference Engines", "level": 4, "role": "Datacenter Inference"},
-    "Desktop Agent Core": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Desktop Agent Core"},
-    "Desktop GUI": {"layer": "Human Interface & System Operations", "level": 10, "role": "Desktop GUI"},
-    "Distributed Search": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Distributed Search"},
-    "Document Archiver": {"layer": "Human Interface & System Operations", "level": 10, "role": "Document Archiver"},
-    "Document Comprehension": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Document Comprehension"},
-    "Document Parser": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Document Parser"},
-    "ETL Transformation": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "ETL Transformation"},
-    "Ecosystem Launcher": {"layer": "Human Interface & System Operations", "level": 10, "role": "Ecosystem Launcher"},
-    "Enterprise Assistant": {"layer": "Human Interface & System Operations", "level": 10, "role": "Enterprise Assistant"},
-    "Extensible Interface": {"layer": "Human Interface & System Operations", "level": 10, "role": "Extensible Interface"},
-    "File Discovery": {"layer": "Development Runtime & Environment", "level": 2, "role": "File Discovery"},
-    "GGUF Runtime": {"layer": "Local Inference Engines", "level": 4, "role": "GGUF Runtime"},
-    "GPU Computing": {"layer": "GPU Acceleration & Optimization", "level": 3, "role": "GPU Computing"},
-    "Gap Analyzer": {"layer": "Human Interface & System Operations", "level": 10, "role": "Gap Analyzer"},
-    "Graph Database": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Graph Database"},
-    "Hardware Acceleration": {"layer": "GPU Acceleration & Optimization", "level": 3, "role": "Hardware Acceleration"},
-    "High-Performance Vector": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "High-Performance Vector"},
-    "IaC DRY Wrapper": {"layer": "Human Interface & System Operations", "level": 10, "role": "IaC DRY Wrapper"},
-    "Image WebUI": {"layer": "Human Interface & System Operations", "level": 10, "role": "Image WebUI"},
-    "Infrastructure Provision": {"layer": "Human Interface & System Operations", "level": 10, "role": "Infrastructure Provision"},
-    "Infrastructure as Code": {"layer": "Development Runtime & Environment", "level": 2, "role": "Infrastructure as Code"},
-    "Integration Library": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Integration Library"},
-    "Isolation": {"layer": "Host Platform & Infrastructure", "level": 1, "role": "Isolation"},
-    "LLM Observability": {"layer": "Human Interface & System Operations", "level": 10, "role": "LLM Observability"},
-    "Layer-wise Inference": {"layer": "Local Inference Engines", "level": 4, "role": "Layer-wise Inference"},
-    "Learning Cards": {"layer": "Human Interface & System Operations", "level": 10, "role": "Learning Cards"},
-    "Lexical Search": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Lexical Search"},
-    "Library Manager": {"layer": "Human Interface & System Operations", "level": 10, "role": "Library Manager"},
-    "Load Balancer": {"layer": "Intelligent API Routers & Proxies", "level": 5, "role": "Load Balancer"},
-    "Local Central Shell": {"layer": "Human Interface & System Operations", "level": 10, "role": "Local Central Shell"},
-    "Local Declarative DSL": {"layer": "Human Interface & System Operations", "level": 10, "role": "Local Declarative DSL"},
-    "Local LLM Runner": {"layer": "Local Inference Engines", "level": 4, "role": "Local LLM Runner"},
-    "Markdown Converter": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Markdown Converter"},
-    "Metal Provisioner": {"layer": "Human Interface & System Operations", "level": 10, "role": "Metal Provisioner"},
-    "Metric Scraper": {"layer": "Human Interface & System Operations", "level": 10, "role": "Metric Scraper"},
-    "Mixed Precision": {"layer": "GPU Acceleration & Optimization", "level": 3, "role": "Mixed Precision"},
-    "Model Curation": {"layer": "Local Inference Engines", "level": 4, "role": "Model Curation"},
-    "Model Manager GUI": {"layer": "Human Interface & System Operations", "level": 10, "role": "Model Manager GUI"},
-    "Model Optimization": {"layer": "GPU Acceleration & Optimization", "level": 3, "role": "Model Optimization"},
-    "Model Surgery": {"layer": "Local Inference Engines", "level": 4, "role": "Model Surgery"},
-    "Multi-Agent Framework": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Multi-Agent Framework"},
-    "Native Inference": {"layer": "Local Inference Engines", "level": 4, "role": "Native Inference"},
-    "Offline Journal": {"layer": "Human Interface & System Operations", "level": 10, "role": "Offline Journal"},
-    "Open-Source IaC": {"layer": "Human Interface & System Operations", "level": 10, "role": "Open-Source IaC"},
-    "Outliner Graph": {"layer": "Human Interface & System Operations", "level": 10, "role": "Outliner Graph"},
-    "Output Evaluator": {"layer": "Human Interface & System Operations", "level": 10, "role": "Output Evaluator"},
-    "PDF Processing": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "PDF Processing"},
-    "Parsing Engine": {"layer": "Development Runtime & Environment", "level": 2, "role": "Parsing Engine"},
-    "Performance Engine": {"layer": "Local Inference Engines", "level": 4, "role": "Performance Engine"},
-    "Platform Workspace": {"layer": "Human Interface & System Operations", "level": 10, "role": "Platform Workspace"},
-    "Production Multi-Agent": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Production Multi-Agent"},
-    "Programmable IaC": {"layer": "Human Interface & System Operations", "level": 10, "role": "Programmable IaC"},
-    "Prompt Tester": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Prompt Tester"},
-    "RAG Flow Canvas": {"layer": "Human Interface & System Operations", "level": 10, "role": "RAG Flow Canvas"},
-    "Reference Manager": {"layer": "Human Interface & System Operations", "level": 10, "role": "Reference Manager"},
-    "Requirement Builder": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Requirement Builder"},
-    "Role-Playing Agent": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Role-Playing Agent"},
-    "Runtime Environment": {"layer": "Development Runtime & Environment", "level": 2, "role": "Runtime Environment"},
-    "Runtime Packaging": {"layer": "Host Platform & Infrastructure", "level": 1, "role": "Runtime Packaging"},
-    "Serverless Database": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Serverless Database"},
-    "Single-File Runtime": {"layer": "Local Inference Engines", "level": 4, "role": "Single-File Runtime"},
-    "Skill Auditor": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Skill Auditor"},
-    "Sovereign Graph Editor": {"layer": "Human Interface & System Operations", "level": 10, "role": "Sovereign Graph Editor"},
-    "Sovereign Software Agent": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Sovereign Software Agent"},
-    "Speech Transcriber": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Speech Transcriber"},
-    "Sprint Planner": {"layer": "Human Interface & System Operations", "level": 10, "role": "Sprint Planner"},
-    "Sprints Manager": {"layer": "Human Interface & System Operations", "level": 10, "role": "Sprints Manager"},
-    "Stateless Orchestration": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Stateless Orchestration"},
-    "Structured Synthesis": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Structured Synthesis"},
-    "Studio Canvas": {"layer": "Human Interface & System Operations", "level": 10, "role": "Studio Canvas"},
-    "Synchronization Layer": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Synchronization Layer"},
-    "System Auditor": {"layer": "Human Interface & System Operations", "level": 10, "role": "System Auditor"},
-    "System Console": {"layer": "Human Interface & System Operations", "level": 10, "role": "System Console"},
-    "Telemetry Monitor": {"layer": "Human Interface & System Operations", "level": 10, "role": "Telemetry Monitor"},
-    "Telemetry Visualizer": {"layer": "Human Interface & System Operations", "level": 10, "role": "Telemetry Visualizer"},
-    "Terminal Agent": {"layer": "Multi-Agent Orchestration Runtimes", "level": 6, "role": "Terminal Agent"},
-    "Terminal Coding Agent": {"layer": "Agentic Software Engineering & Sandboxes", "level": 7, "role": "Terminal Coding Agent"},
-    "Text Search": {"layer": "Development Runtime & Environment", "level": 2, "role": "Text Search"},
-    "Unified API Gateway": {"layer": "Intelligent API Routers & Proxies", "level": 5, "role": "Unified API Gateway"},
-    "Vector Engine": {"layer": "Decentralized Knowledge & Vector Stores", "level": 8, "role": "Vector Engine"},
-    "Vision describer": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Vision describer"},
-    "Visual Agent Canvas": {"layer": "Human Interface & System Operations", "level": 10, "role": "Visual Agent Canvas"},
-    "Voice Synthesis": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Voice Synthesis"},
-    "Web Scraper": {"layer": "Data Extraction & Pipeline Harvest", "level": 9, "role": "Web Scraper"},
-}
-
-def migrate_constants(path: Path):
-    if not path.exists():
-        print(f"[-] constants.py not found at {path}")
-        return
-    content = path.read_text(encoding="utf-8")
-    
-    # Locate NAV_LAYER_ORDER array
-    nav_pattern = re.compile(r'NAV_LAYER_ORDER:\s*list\[str\]\s*=\s*\[[^\]]+\]', re.DOTALL)
-    new_nav_str = "NAV_LAYER_ORDER: list[str] = [\n" + "".join(f'    "{l}",\n' for l in NEW_LAYERS) + "]"
-    
-    if nav_pattern.search(content):
-        content = nav_pattern.sub(new_nav_str, content)
-        path.write_text(content, encoding="utf-8")
-        print(f"[+] Migrated constants.py layers!")
-    else:
-        print(f"[-] NAV_LAYER_ORDER block not matched in constants.py")
-
-def migrate_db_manager(path: Path):
-    if not path.exists():
-        print(f"[-] db_manager.py not found at {path}")
-        return
-    content = path.read_text(encoding="utf-8")
-    
-    # Locate CATEGORY_MAP dict block
-    map_pattern = re.compile(r'CATEGORY_MAP:\s*dict\[str,\s*dict\[str,\s*object\]\]\s*=\s*\{[^}]+(\}[^}]*)*\}', re.DOTALL)
-    
-    # Format CATEGORY_MAP dict
-    new_map_str = "CATEGORY_MAP: dict[str, dict[str, object]] = {\n"
-    for cat in sorted(NEW_CATEGORY_MAP.keys()):
-        val = NEW_CATEGORY_MAP[cat]
-        new_map_str += f'    "{cat}": ' + "{" + f'"layer": "{val["layer"]}", "level": {val["level"]}, "role": "{val["role"]}"' + "},\n"
-    new_map_str += "}"
-    
-    if map_pattern.search(content):
-        content = map_pattern.sub(new_map_str, content)
-        path.write_text(content, encoding="utf-8")
-        print(f"[+] Migrated db_manager.py CATEGORY_MAP!")
-    else:
-        print(f"[-] CATEGORY_MAP block not matched in db_manager.py")
-
-def migrate_connections(path: Path, defaults_path: Path):
-    if not path.exists():
-        print(f"[-] connections.py not found at {path}")
-        return
-    if not defaults_path.exists():
-        print(f"[-] defaults.py not found at {defaults_path} - cannot resolve dynamic tool layers")
-        return
-        
-    # Dynamically read tool_id -> layer from defaults.py
-    sys.path.insert(0, str(defaults_path.parent))
-    import defaults
-    tool_to_layer = {}
-    for tid, tool in defaults.DEFAULT_REGISTRY.items():
-        tool_to_layer[tid] = tool["layer"]
-        
-    content = path.read_text(encoding="utf-8")
-    
-    # Add DEFAULT_REGISTRY import to connections.py if missing
-    import_line = "from ai_lsc.registry.defaults import DEFAULT_REGISTRY"
-    if "import DEFAULT_REGISTRY" not in content and import_line not in content:
-        # Insert after absolute imports
-        content = "from ai_lsc.registry.defaults import DEFAULT_REGISTRY\n" + content
-        print("[+] Added DEFAULT_REGISTRY import to connections.py")
-
-    # Reorganize the StackWiring layer arguments in connections.py
-    blocks = content.split("_reg(StackWiring(")
-    new_blocks = [blocks[0]]
-    updated_count = 0
-    
-    for block in blocks[1:]:
-        tid_match = re.search(r"""tool_id\s*=\s*["']?([a-zA-Z0-9_-]+)["']?""", block)
-        if tid_match:
-            tid = tid_match.group(1)
-            if tid in tool_to_layer:
-                new_layer = tool_to_layer[tid]
-                # Replace layer="..." or layer='...'
-                block, count = re.subn(r"""layer\s*=\s*["|'][^"'\s]+["|']""", f'layer="{new_layer}"', block)
-                if count:
-                    updated_count += 1
-        new_blocks.append(block)
-        
-    content = "_reg(StackWiring(".join(new_blocks)
-    
-    # Convert hardcoded loops to look up the layer from the registry dynamically
-    # Example: layer="DevOps" inside loops -> layer=DEFAULT_REGISTRY[_tid]["layer"]
-    content, loops_updated = re.subn(
-        r"""layer\s*=\s*["'](?:DevOps|User Interfaces|Knowledge Management|Orchestrators|Host Platform|Development Environment|GPU Runtimes|Engines|Routing|Security|Observability)["']""",
-        'layer=DEFAULT_REGISTRY[_tid]["layer"]',
-        content
-    )
-    if loops_updated:
-        print(f"[+] Converted {loops_updated} loop-based StackWiring layers to dynamic lookups!")
-        
-    path.write_text(content, encoding="utf-8")
-    print(f"[+] Migrated connections.py! Rebuilt {updated_count} static tool-layer definitions.")
-
-def migrate_layers(layers_dir: Path, defaults_path: Path):
-    if not layers_dir.is_dir():
-        print(f"[-] Layers directory not found at {layers_dir}")
-        return
-    if not defaults_path.exists():
-        print(f"[-] defaults.py not found - cannot update layer files")
-        return
-        
-    import defaults
-    tool_to_layer = {}
-    tool_to_level = {}
-    for tid, tool in defaults.DEFAULT_REGISTRY.items():
-        tool_to_layer[tid] = tool["layer"]
-        tool_to_level[tid] = tool["level"]
-        
-    updated_files = 0
-    for path in layers_dir.glob("*.py"):
-        if path.name == "__init__.py":
-            continue
-        original = path.read_text(encoding="utf-8")
-        content = original
-        
-        # Replace level and layer for each tool inside the file
-        for tid in tool_to_layer:
-            # Look for block: 'tool_id': { or "tool_id": {
-            # we do a specific match and replace within that dict
-            tool_pattern = re.compile(rf"""["']{tid}["']\s*:\s*\{{[^}}]+}}""", re.DOTALL)
-            match = tool_pattern.search(content)
-            if match:
-                block = match.group(0)
-                # Replace level
-                block = re.sub(r'"level":\s*\d+', f'"level": {tool_to_level[tid]}', block)
-                # Replace layer
-                block = re.sub(r""""layer":\s*["|'][^"'\s]+["|']""", f'"layer": "{tool_to_layer[tid]}"', block)
-                content = content.replace(match.group(0), block)
-                
-        if content != original:
-            path.write_text(content, encoding="utf-8")
-            print(f"[+] Realigned tool layers & levels inside modular layer file: {path.name}")
-            updated_files += 1
-            
-    print(f"[+] Layer files update complete. Modified {updated_files} layer modules.")
-
-def main():
-    import sys
-    root = Path(__file__).resolve().parent
-    src_dir = root / "src"
-    
-    # Setup paths relative to root or src
-    constants_path = src_dir / "ai_lsc" / "constants.py"
-    db_manager_path = src_dir / "ai_lsc" / "ui" / "pages" / "db_manager.py"
-    connections_path = src_dir / "ai_lsc" / "stack" / "connections.py"
-    defaults_path = src_dir / "ai_lsc" / "registry" / "defaults.py"
-    layers_dir = src_dir / "ai_lsc" / "registry" / "layers"
-    
-    print("[*] Starting taxonomy migration sweeps...")
-    migrate_constants(constants_path)
-    migrate_db_manager(db_manager_path)
-    migrate_connections(connections_path, defaults_path)
-    migrate_layers(layers_dir, defaults_path)
-    print("[+] Entire codebase has been successfully upgraded to the 10-Layer Architecture!")
-
-if __name__ == "__main__":
-    main()

src/ai_lsc/registry/defaults.py

@@ -3132,7 +3132,7 @@ DEFAULT_REGISTRY: dict = {
     },
     "launcher": {
         "type": 'tmux',
-        "cmd": 'open-webui serve --port {port} --data-dir {workspaces_root}/openwebui',
+        "cmd": 'WEBUI_DATA_DIR={workspaces_root}/openwebui open-webui serve --port {port}',
         "default_port": 8080,
     },
     "deps": [

src/ai_lsc/registry/layers/user_interfaces.py

@@ -24,7 +24,7 @@ TOOLS: dict[str, dict] = {
     },
     "launcher": {
         "type": "tmux",
-        "cmd": "open-webui serve --port {port} --data-dir {workspaces_root}/openwebui",
+        "cmd": "WEBUI_DATA_DIR={workspaces_root}/openwebui open-webui serve --port {port}",
         "default_port": 8080
     },
     "deps": [
@@ -540,7 +540,7 @@ TOOLS: dict[str, dict] = {
     },
     "launcher": {
         "type": "tmux",
-        "cmd": "open-webui serve --port {port} --data-dir {workspaces_root}/hermes-webui --env OLLAMA_BASE_URL=http://localhost:17051",
+        "cmd": "WEBUI_DATA_DIR={workspaces_root}/hermes-webui OLLAMA_BASE_URL=http://localhost:17051 open-webui serve --port {port}",
         "default_port": 8081
     },
     "deps": [

src/ai_lsc/stack/connections.py

@@ -3398,13 +3398,13 @@ _reg(StackWiring(
                 decision="Run a second Open-WebUI instance dedicated "
                          "to the Hermes stack.",
                 observation="Open-WebUI supports multiple instances "
-                            "with separate data-dir flags. The general "
+                            "with separate WEBUI_DATA_DIR env vars. The "
-                            "openwebui tool already occupies :3000 with "
+                            "general openwebui tool already occupies :3000 "
-                            "Ollama direct as backend.",
+                            "with Ollama direct as backend.",
                 alternatives="Use a single Open-WebUI instance with "
                              "model name prefixes (hermes/*).",
                 constraints="Must run on a different port (8081) and "
-                            "data-dir than the general openwebui.",
+                            "WEBUI_DATA_DIR than the general openwebui.",
                 reasoning="Dedicated Hermes UI keeps the Hermes agent "
                           "runtime as the single backend, so every "
                           "Hermes WebUI conversation flows through "