@futdevpro/fsm-dynamo 1.16.37 → 1.16.39

package/__documentations/2026-05-17-oai-compatible-providers-howto.md

@@ -1,282 +0,0 @@
-# 2026-05-17 — OpenAI-compatible providers (Hunglish how-to)
-
-**Spec:** `__agent/feature-requests/FR-005-oai-compat-provider-docs.md` (workspace root)
-**Scope:** docs-only, LOW priority. Nincs API-változás.
-
----
-
-## Mire jó ez?
-
-A `DyFM_OAI_ClientOptions.baseURL` mező (`fsm-dynamo/_modules/ai/_modules/open-ai/_models/oai-client-options.interface.ts:25`) lehetővé teszi **tetszőleges OpenAI-API-kompatibilis endpoint** használatát az `openai` npm SDK-n keresztül — anélkül, hogy a Dynamo OAI service-eket (`DyFM_OAI_Settings`, downstream `DyNTS_OAI_Embedding_ControlService` / `DyNTS_OAI_LLM_ServiceBase`) bármi módon kellene módosítani.
-
-Tipikus use-case-ek:
-- **Lokál fejlesztés**: `OPENAI_API_KEY` nélkül, lokál LLM-mel (LM Studio, Ollama)
-- **Self-hosted production**: vLLM / LocalAI klaszter saját infrán
-- **Cost-control**: nyílt-súlyú modellek (Llama, Mistral, Qwen) drága API helyett
-
-A `baseURL`-override `openai` SDK-natív feature — Dynamo csak átadja a config-ot (`oai-llm.service-base.ts:91-98`):
-
-```typescript
-this.openai = new OpenAI(
-  set?.config ?? {
-    organization: DyNTS_global_settings.env_settings.openAi.organization,
-    apiKey: DyNTS_global_settings.env_settings.openAi.apiKey,
-    project: DyNTS_global_settings.env_settings.openAi.project,
-  }
-);
-```
-
-→ ha a `set.config.baseURL` ki van töltve, az SDK arra megy. Minden más változatlan.
-
----
-
-## Általános minta
-
-```typescript
-import { DyFM_OAI_Settings, DyFM_OAI_CallSettings } from '@futdevpro/fsm-dynamo/ai/open-ai';
-import { DyNTS_OAI_LLM_ChatServiceBase } from '@futdevpro/dynamo-nts/ai/open-ai';
-
-const settings: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: {
-    baseURL: 'http://<provider-host>:<port>/v1',
-    apiKey: '<placeholder-or-real>',
-  },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: '<provider-specific-model-id>',
-  }),
-});
-
-// Use with any Dynamo OAI service:
-class MyChat extends DyNTS_OAI_LLM_ChatServiceBase {}
-const chat: MyChat = new MyChat(settings);
-```
-
-A kulcs: `config.baseURL` + `config.apiKey` + `defaultSettings.useModel`. A többi mező (organization, project) lokál provider-eknél figyelmen kívül marad.
-
----
-
-## Provider-szekciók
-
-### 1. LM Studio — desktop GUI lokál LLM-hez
-
-[LM Studio](https://lmstudio.ai) Mac/Windows/Linux desktop app, beépített OAI-compat szerverrel. Default port: `1234`.
-
-**Beállítás**:
-1. LM Studio-ban betölteni egy modellt (pl. `nomic-embed-text-v1.5` embedding-hez, `llama-3.2-3b-instruct` chat-hez)
-2. "Local Server" tab → "Start Server"
-
-**Code**:
-```typescript
-const lmStudio: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: {
-    baseURL: 'http://localhost:1234/v1',
-    apiKey: 'lm-studio',  // tetszőleges placeholder — LM Studio ignorálja
-  },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: 'nomic-embed-text-v1.5',  // ahogy a Local Server megjeleníti
-  }),
-});
-```
-
-**Tipikus jellemzők**:
-- ✅ GUI a modell-letöltéshez (HuggingFace integráció)
-- ✅ Streaming chat completion support
-- ⚠️ Embedding limit: a UI-n külön kell betölteni embedding modellt
-- ❌ Multi-tenant / production-flow nem támogatott
-
----
-
-### 2. Ollama — CLI-driven lokál LLM (OAI-compat layer ≥ v0.1.14)
-
-[Ollama](https://ollama.com) parancssoros LLM-runner, OAI-compat layerrel. Default port: `11434`.
-
-**Beállítás**:
-```bash
-ollama pull llama3.2:3b
-ollama pull nomic-embed-text
-ollama serve  # vagy ha service-ként fut, már megy
-```
-
-**Code**:
-```typescript
-const ollama: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: {
-    baseURL: 'http://localhost:11434/v1',
-    apiKey: 'ollama',  // szintén placeholder — Ollama nem ellenőrzi
-  },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: 'nomic-embed-text',
-  }),
-});
-```
-
-**Tipikus jellemzők**:
-- ✅ Multi-modell support (`ollama pull <name>` → azonnal elérhető endpointon)
-- ✅ Embedding endpoint OAI-compat formában működik (v0.1.14+ óta)
-- ✅ Streaming chat completion OK
-- ⚠️ `function_call` / `tools` API csak részben támogatott a tool-aware modelleknél (Llama-3 family OK; small instruct modellek may not)
-- ⚠️ Context-window provider-szintű limit (általában 2048-8192 tokens default-tal — `OLLAMA_NUM_CTX` env-var)
-
----
-
-### 3. vLLM — production-grade GPU inference server
-
-[vLLM](https://docs.vllm.ai) Python alapú, batched + PagedAttention GPU-inference. OAI-compatible REST endpoint beépítve. Default port: `8000`.
-
-**Beállítás**:
-```bash
-pip install vllm
-python -m vllm.entrypoints.openai.api_server \
-  --model meta-llama/Meta-Llama-3-8B-Instruct \
-  --port 8000 \
-  --api-key local-secret-token
-```
-
-**Code**:
-```typescript
-const vllm: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: {
-    baseURL: 'http://gpu-server:8000/v1',
-    apiKey: process.env.VLLM_API_KEY ?? 'local-secret-token',  // ha --api-key beállítva
-  },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: 'meta-llama/Meta-Llama-3-8B-Instruct',  // a `--model` érték
-  }),
-});
-```
-
-**Tipikus jellemzők**:
-- ✅ Magas throughput (batched inference) — production scaling
-- ✅ Tool-calling support latest verziókban
-- ✅ Streaming OK
-- ⚠️ Egyszerre **egy** modell van memóriában (több modell = több instance)
-- ⚠️ Embedding endpoint NEM all modellnél — csak ha a model dedikált embedding (pl. `intfloat/e5-mistral-7b-instruct`)
-- ⚠️ GPU-szintű mem-management — OOM-ra dob, ha context-len túl nagy
-
----
-
-### 4. LocalAI — REST-only multi-model platform
-
-[LocalAI](https://localai.io) Go-alapú, multi-backend (llama.cpp, whisper.cpp, stable-diffusion) REST szerver. OAI-compatible chat + embedding endpointokkal. Default port: `8080`.
-
-**Beállítás**:
-```bash
-docker run -p 8080:8080 \
-  -v $PWD/models:/build/models \
-  localai/localai:latest
-```
-
-Modellt vagy `models.yaml`-ban vagy runtime-on (`POST /models/apply`) lehet hozzáadni.
-
-**Code**:
-```typescript
-const localai: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: {
-    baseURL: 'http://localai:8080/v1',
-    apiKey: 'sk-local',  // placeholder
-  },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: 'gpt-4',  // LocalAI alias → konfigban mapped a tényleges modellre
-  }),
-});
-```
-
-**Tipikus jellemzők**:
-- ✅ Egyszerre több modell loadolva — modell-aliasok
-- ✅ Multi-backend (LLM + Whisper transcribe + image-gen ugyanazon szerverről)
-- ✅ Streaming OK
-- ⚠️ A modell-aliasok kézi konfig kérdése — a `useModel` ID-nek pontosan match-elnie kell a `models.yaml`-ban definiálttal
-- ⚠️ Lassabb felpörgés gpu-acceleration nélkül (CPU-fallback default)
-
----
-
-## Embedding modellek — provider-specifikus
-
-Az embedding-modell nevének **pontos egyezést** kell mutatnia a provider által regisztrált name-mel. Példák:
-
-| Provider | Embedding model ID | Megjegyzés |
-|---|---|---|
-| OpenAI (cloud) | `text-embedding-3-small`, `text-embedding-3-large` | Default |
-| LM Studio | `nomic-embed-text-v1.5`, `BAAI/bge-large-en-v1.5` | A betöltött modell ID-jét a Local Server tab írja ki |
-| Ollama | `nomic-embed-text`, `mxbai-embed-large` | `ollama list` mutatja |
-| vLLM | `intfloat/e5-mistral-7b-instruct` | Csak ha az indítási `--model` embedding-képes |
-| LocalAI | `text-embedding-ada-002` (aliasként) | A `models.yaml`-ban definiált alias |
-
-A `DyNTS_OAI_Embedding_ControlService` API-szintjén nem érzékeli a különbséget — a `useModel` field határoz meg mindent.
-
----
-
-## Caveats — amire figyelni kell
-
-### Rate limits
-- **OpenAI**: tier-szintű, API-doc-ban
-- **LM Studio / Ollama / LocalAI**: nincs natív rate-limit — single-user dev tool
-- **vLLM**: konfigurálható (`--max-num-seqs` flag), default kvázi-unlimited
-
-### Max tokens / context window
-A `DyFM_OAI_CallSettings.maxTokens` provider-szintű limit alá kell hogy essen, különben az SDK 400/422-t kap:
-
-| Provider | Default context-window | Override |
-|---|---|---|
-| OpenAI gpt-4o | 128K | API tier-függő |
-| LM Studio | a betöltött modell native context-je | UI slider |
-| Ollama | 2048 (legacy default!) → `num_ctx` paraméter | `OLLAMA_NUM_CTX` env-var vagy `Modelfile` `PARAMETER num_ctx` |
-| vLLM | a modell native context-je | `--max-model-len` flag |
-| LocalAI | `models.yaml` `context_size` field | per-model konfig |
-
-### Error-handling különbségek
-- **OpenAI**: standard 400/401/403/429/500 + structured error body
-- **Lokál providerek**: gyakran **404 a modell-name-re** (ha rosszul ID-zed), **timeout** (modell-loadolás közben), **OOM** (vLLM)
-- **Ollama-specifikus**: ha a modell még nem `ollama pull`-olt, az endpoint csendben várakozik a pull-ig (lassú first-call)
-- **LocalAI**: model-name mismatch → 422-t dob OAI-szerű error-body-val
-
-### Function-calling / tool-use
-- **OpenAI**: teljes function-calling support
-- **Ollama**: model-függő — Llama-3.1+ és Mistral tool-aware modellek igen, kicsi instruct modellek nem
-- **vLLM**: support latest (v0.6+), de chat-template-mappel paraméterezni kell
-- **LM Studio / LocalAI**: parciális, model-függő
-
-### Streaming
-Mind a 4 provider támogat OAI-compat SSE streaming-et, **de**:
-- Ollama legacy verziók (< v0.1.20) `stream: false`-cal stabilabb
-- LocalAI streaming-rate alacsonyabb mint a token-emission
-
----
-
-## Quick recipe — Dynamo-szintű váltás dev → cloud
-
-```typescript
-const isDev: boolean = process.env.NODE_ENV === 'development';
-
-const aiSettings: DyFM_OAI_Settings = new DyFM_OAI_Settings({
-  config: isDev
-    ? {
-        baseURL: 'http://localhost:11434/v1',  // Ollama
-        apiKey: 'dev',
-      }
-    : {
-        apiKey: process.env.OPENAI_API_KEY,
-        organization: process.env.OPENAI_ORG_ID,
-      },
-  defaultSettings: new DyFM_OAI_CallSettings({
-    useModel: isDev ? 'llama3.2:3b' : 'gpt-4o',
-  }),
-});
-```
-
-A Dynamo OAI service-ek mindkét beállítással ugyanúgy működnek.
-
----
-
-## Hivatkozott source-fájlok
-
-| Fájl | Mit ad |
-|---|---|
-| `fsm-dynamo/src/_modules/ai/_modules/open-ai/_models/oai-client-options.interface.ts` | `DyFM_OAI_ClientOptions.baseURL` mező |
-| `fsm-dynamo/src/_modules/ai/_modules/open-ai/_models/oai-settings.control-model.ts` | `DyFM_OAI_Settings` wrapper |
-| `dynamo-nts/src/_modules/ai/_modules/open-ai/_services/oai-llm.service-base.ts` | OpenAI SDK instantiation (config átadás) |
-| `dynamo-nts/src/_modules/ai/_modules/open-ai/_services/oai-embedding.control-service.ts` | Embedding endpoint usage |
-
-## Backward compatibility
-
-A `baseURL` mező mindig is létezett az OpenAI SDK-ban — a Dynamo egyetlen változtatása ehhez a docs-ot tisztázza, nem feature.

package/__documentations/2026-06-01-fr047-fc-foundation.md

@@ -1,48 +0,0 @@
-# FR-047 — function-calling foundation in dynamo-fsm (2026-06-01)
-
-**Parent**: `__agent/feature-requests/FR-047-provider-agnostic-function-calling.md` (+ Phase 1) and `FR-048-per-model-agnostic-settings.md`
-**Versions**: `01.15.17` (Phase 1 + registry), `01.15.18` (adultContent), `01.15.19` (export-bug fix)
-**Scope**: `src/_modules/ai` (agnostic core + provider model submodules), `package.json` (exports)
-
-## What
-
-The SDK-free agnostic AI core gained the full function-calling vocabulary + a per-model capability registry, plus a reserved per-model settings extension point. No execution logic here — that lives in dynamo-nts (Phase 2+).
-
-### 01.15.17 — Phase 1 vocabulary + registry foundation + FR-048 reservation
-- Tool models: `DyFM_AI_Tool`, `DyFM_AI_ToolCall`, `DyFM_AI_ToolResult`, `DyFM_AI_ToolHandler`/`DyFM_AI_ToolHandlers`.
-- `DyFM_AI_Message` + `DyFM_AI_LLM_Response` gained optional tool fields (`toolCalls`, `toolCallId`, `stopReason`); `DyFM_AI_MessageRole.tool` added.
-- `DyFM_AI_ModelType` extended: `Image`, `Audio`.
-- Per-model registry: `DyFM_AI_ModelInfo` + `DyFM_AI_ModelCapabilities` + the 5 provider const arrays (`DyFM_OAI_Models`, `DyFM_AAI_Models`, `DyFM_GAI_Models`, `DyFM_LAI_Models`, `DyFM_FdpAI_Models`) — verified 2026-06-01 against official provider docs.
-- `DyFM_AI_ModelRegistry_Util` (`findModelInfo`, `modelSupportsTools`) — takes the registry array as a param so the agnostic core imports NO provider submodule (keeps it SDK-free).
-- FR-048: `DyFM_AI_ModelSettingsSchema` reserved field on `DyFM_AI_ModelInfo` + thorough docs (per-model settings system; NOT implemented).
-
-### 01.15.18 — adultContent capability
-- `DyFM_AI_ModelCapabilities.adultContent: boolean` — can the model generate adult / uncensored / NSFW content. For content-routing + age-gating (e.g. Adventor mature mode). All 21 registered models = `false` (mainstream cloud is content-filtered; local/FdpAI conservative / model-dependent).
-
-### 01.15.19 — subpath export-bug fix
-- `package.json` `exports` + `typesVersions`: `./ai/google` and `./ai/local` pointed at non-existent `_modules/ai/_modules/google`/`local` folders; the real folders are `google-ai`/`local-ai`. Fixed the targets → `@futdevpro/fsm-dynamo/ai/google` and `./ai/local` now resolve. Unblocks dynamo-nts Phase 5 (Google).
-
-## Why
-
-Function calling needed to live in the strategic provider-agnostic layer (not only the legacy OpenAI-specific `FDPNTS_GPT_ControlService`). dynamo-fsm owns the vendor-neutral vocabulary + the model registry; dynamo-nts owns the execution (agent-loop + per-provider translators). This commit set delivers the fsm half so nts can build against it.
-
-## Design notes
-
-- All changes additive/non-breaking except the export-path fix (corrects a latent resolution bug) and the new required `adultContent` capability (all registry consts updated in the same release).
-- Capability honesty: `functionCalling` (and `adultContent`) are per-model; unknown/local models default conservative (`false`) — never assume capability.
-- Tool-handler runner semantics (never-throw, errors returned as strings to the model) are defined by the type + implemented in dynamo-nts.
-
-## Tests
-
-`pnpm test` (build-base + jasmine): **1302 specs / 0 failures / 70 pending** across the releases. New: `ai-model-registry.util.spec.ts` (findModelInfo + modelSupportsTools, incl. unknown-id → false).
-
-## What's NOT here
-
-- The agent-loop + provider translators (`requestWithTools`, `runToolLoop`, OpenAI/Anthropic/Gemini tool translators) — dynamo-nts (FR-047 Phase 2–5).
-- The per-model settings reconciler — FR-048 (reserved only).
-
-## Cross-references
-
-- `__agent/feature-requests/FR-047-*.md`, `FR-048-per-model-agnostic-settings.md`
-- `fdp-documentations/guidelines/development/ai-per-model-settings.md`
-- dynamo-nts Phase 2+3 (agent-loop + OpenAI translator) — implemented in parallel.

package/__documentations/2026-06-01-fr047-fr048-code-review.md

@@ -1,153 +0,0 @@
-# 2026-06-01 — Code Review: FR-047 + FR-048 (provider-agnostic AI vocabulary)
-
-## Summary
-
-Code review futott a `dynamo-fsm` master branchen az utolsó 3 commit-ra (FR-047 + FR-048) + 1 uncommitted root pipeline-config deletion-re. Az effort-szint **high**: 7 finder angle × ≤6 candidate → 1-vote verify (recall-biased) → top-10 finding.
-
-### Scope
-
-- **Range**: `HEAD~3..HEAD` (master, no upstream divergence)
-- **Commits**:
-  - `08e8859` feat(ai): FR-047 P1 provider-agnostic function-calling vocabulary + per-model registry; FR-048 settings-schema reservation
-  - `de1da9b` feat(ai): add adultContent (uncensored/NSFW) per-model capability
-  - `6a4ce9c` fix(ai): FR-047 P5 — correct ./ai/google + ./ai/local subpath export targets
-- **Uncommitted**: `pipeline.cicd.config.json` (root) -128 sor delete (dmc migracio takaritas)
-- **Source-diff stat**: 30 fajl, 940 inserciol / 12 deletes (build/ artifact kizarva)
-
-### Tool
-
-A `/code-review` skill (high effort), local execution. Phase 1 — 7 párhuzamos finder agent (Explore type):
-- **3 correctness angle**: line-by-line scan, removed-behavior auditor, cross-file tracer
-- **3 cleanup angle**: reuse, simplification, efficiency
-- **1 altitude angle**: depth check
-
-Phase 2 — 1-vote verifier batch agent (recall-biased), 17 candidate → 16 survivor (1 REFUTED).
-
-## Top-10 Finding (ranked most-severe first)
-
-A teljes JSON output `__documentations/2026-06-01-fr047-fr048-code-review-findings.json`-ben. Itt a Hunglish summary:
-
-### 1. `gptImage_1` enum-entry registry-mentes (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_modules/open-ai/_enums/oai-model.enum.ts:19`
-
-A `DyFM_OAI_Model.gptImage_1 = 'gpt-image-1'` enum-ben szerepel + komment szerint **"current"** (nem deprecated), de a `DyFM_OAI_Models` registry-ben NEM kap entry-t.
-
-**Failure scenario**: `modelSupportsTools(DyFM_OAI_Models, DyFM_OAI_Model.gptImage_1)` → false (a util `?? false`-szal default-ol). `findModelInfo` undefined. Az image generation tools-on-keresztul rutalas silent-skip.
-
-### 2. `claude_3_5_sonnet_latest` enum + nincs registry (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_modules/anthropic/_enums/aai-model.enum.ts:11`
-
-A `DyFM_AAI_Models` registry csak 3 GA entry-t tartalmaz; a deprecated `claude_3_5_sonnet_latest` az enum-ben van, registry-bol kimaradt.
-
-**Failure scenario**: Default `DyFM_AI_CallSettings` ami a `claude_3_5_sonnet_latest`-et hasznalja → `modelSupportsTools` false → tool-calling silent-disable egy meg-mindig-elo API modelen.
-
-### 3. `gemini_1_5_pro_latest` enum + nincs registry (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_modules/google-ai/_enums/gai-model.enum.ts:11`
-
-Ugyanaz a minta mint #2 — google oldalon.
-
-### 4. `modelSupportsTools` silent-false unknown-ID-ra (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_collections/ai-model-registry.util.ts:27`
-
-`Array.find()` → `?? false` minta. Tipo, deprecated, vagy live-but-unregistered model esten → `false` warn/log nelkul.
-
-**Failure scenario**: Diagnozis nelkul lehetetlen kideriteni miert nem hivodik a tool-routing. Egy `{ strict: true }` mode vagy `DyFM_Log.warn` flegging hianyzik.
-
-### 5. Legacy + active enum-tagok keveredtek (CONFIRMED)
-
-**Fajlok**: `aai/gai/oai-model.enum.ts`
-
-Csak komment kulonbozteti meg az active es retired model-ID-kat. Az enum public API-jaban mindketto egyformanak latszik. A #1-3 ennek tuneti megnyilvanulasa.
-
-**Tervezet fix**: `*-legacy.enum.ts` kulon fajl, vagy minden retired enum-tag mellé `@deprecated` JSDoc tag.
-
-### 6. `DyFM_AI_Message.content` required, de tool-calls esetn ures (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_models/ai-message.interface.ts:6`
-
-`content: string` (non-optional). Mikor `role='assistant'` + `toolCalls` populalva → a content semantically ures OpenAI/Anthropic/Gemini API-knal.
-
-**Failure scenario**: Caller hard-coded `content: ''` workaround minden tool-call message-en; downstream "content empty" validator visszadobhatja a tool-request message-eket.
-
-### 7. `settingsSchema` ghost contract (CONFIRMED)
-
-**Fajl**: `src/_modules/ai/_models/ai-model-info.interface.ts:37`
-
-`settingsSchema?: DyFM_AI_ModelSettingsSchema` reserved minden modelen, de a 21 registry entry kozul EGYIK SEM populalja.
-
-**Failure scenario**: Phase 2 (FR-048) reconciler reading `settingsSchema` undefined → "no constraints" semantikat olvas a "constraints not yet modeled" helyett → silent-accept `temperature` parameter Opus 4.7+ hivason → HTTP 400 hard-fail.
-
-### 8. `finishReason` vs `stopReason` dual-field (PLAUSIBLE)
-
-**Fajl**: `src/_modules/ai/_models/ai-llm-response.interface.ts:16`
-
-Mindketto optional. Legacy `finishReason` + new `stopReason`. Doc-mentes melyik az authoritative.
-
-**Failure scenario**: Provider translator csak `stopReason`-t populalja → legacy consumer `finishReason === 'tool_use'`-t check-el → condition false → tool flow miss.
-
-### 9. Tool-vocab interfaces 0-caller (PLAUSIBLE)
-
-**Fajl**: `src/_modules/ai/_models/ai-tool.interface.ts`
-
-Uj `DyFM_AI_Tool`, `DyFM_AI_ToolCall`, `DyFM_AI_ToolResult`, `DyFM_AI_ToolHandler` exportalt **dynamo-fsm-bol**, de **ZERO caller** sajat-projekten belul. Az abstrakcio nem validalt.
-
-**Failure scenario**: A shape lehet hogy nem fer ra a 5 provider mind-egyikere (OpenAI `type:'function'` wrapper, Anthropic `input_schema`, Gemini `functionDeclarations`). FR-048 mid-flight API redesign-kenyszer.
-
-### 10. Root pipeline.cicd.config.json -128 sor (CONFIRMED, uncommitted)
-
-**Fajl**: `pipeline.cicd.config.json` (root, delete)
-
-A `.dynamo/pipeline.cicd.config.json` fully fedi, DE barmely external workflow / GitHub Actions / script ami hardcoded `./pipeline.cicd.config.json` ref-fel → 404.
-
-## Levágott (cleanup-class — kovetkezo iteracio)
-
-A skill rule: "Correctness bugs always outrank cleanup and altitude findings when the output cap forces a cut." Ezert a top-10 keret felett kihagyott 6 valid finding:
-
-| # | Fajl | Issue |
-|---|---|---|
-| 11 | `src/_modules/ai/index.ts` | No meta-barrel a 5 provider model-registry-re; Phase 2 hardcodes 5 imports |
-| 12 | `ai-model-capabilities.interface.ts` + 5 const-fajl | Capability matrix copy-paste 21 model × 9 mező — N+1 edit kotelezo |
-| 13 | `ai-model-capabilities.interface.ts:31` | `adultContent` policy-flag feature-capabilities matrix-ban; jovobeli policy flags scatterolnek |
-| 14 | `package.json` exports | Naming inconsistency: `anthropic` (no hyphen) vs `open-ai/google-ai/local-ai/fdp-ai` |
-| 15 | `ai-model-registry.util.ts:17` | O(N) `Array.find` registry lookup vs O(1) Map |
-| 16 | Provider `index.ts` (×5) | Tree-shake loss: barrel pulls all 5 const-tables (~519 lines) for single-provider consumer |
-
-## REFUTED (1 db)
-
-| # | Fajl | Reason |
-|---|---|---|
-| - | `oai-models.const.ts:235` gptImage_2 missing maxOutputTokens | Verified: `maxOutputTokens?` optional a `DyFM_AI_ModelInfo` interface-en, no type-level contract violation, no crash |
-
-## Strukturalis tanulsag (kozos pattern a #1-5 mogott)
-
-A top-10 elso 5 finding mind UGYANARRA a strukturalis hiba-mintara mutat: **az enum-tagok es a registry-entries nincsenek szinkronban**. Tervezet javitas:
-
-**Opció A** (defensive): minden enum-entry-nek legyen registry-entry, deprecated-eket `*-legacy.enum.ts`-be kulonitsd el.
-
-**Opció B** (strict-mode): `modelSupportsTools(models, id, { strict: true })` — unknown ID-ra throw vagy `DyFM_Log.warn`.
-
-**Opció C** (build-time validation): TS guard-test ami compile-time verifikalja hogy minden enum-tag registry-entry-vel jar.
-
-Mind a 3 megoldhato — a #1-5 + #11 (meta-barrel) kombo eseten egy follow-up commit lefedi.
-
-## Stat
-
-| | |
-|---|---|
-| **Specs** | 1302 spec / 0 failure (TypeScript + jasmine) |
-| **TS compile** | ✓ (0 error) |
-| **Diff scope** | 30 src fajl / 940 ins / 12 del (build/ kizarva) |
-| **Finder agents** | 7 (parallel) |
-| **Verifier agents** | 1 (batch over 17 candidates) |
-| **Output cap** | 10 finding (skill default) |
-| **Time** | ~15 min full pipeline |
-
-## Reference
-
-- Skill: `code-review` (high effort)
-- Skill instrukciok: 3 correctness + 3 cleanup + 1 altitude angle × ≤6 candidates per angle; 1-vote verify per candidate; recall-biased PLAUSIBLE default; REFUTED only constructible from code
-- Workflow: lokal jasmine `npx jasmine --config=spec/support/jasmine.json` → 1302/0

package/__documentations/2026-06-01-fr047-p1-function-calling-vocabulary.md

@@ -1,75 +0,0 @@
-# FR-047 P1 — Provider-agnostic function-calling vocabulary + per-model registry (2026-06-01)
-
-**Commit**: `08e8859` — v01.15.16 → v01.15.17 (published)
-**Parent**: `__agent/feature-requests/FR-047-phase-1-tool-abstraction.md` (master: `FR-047-provider-agnostic-function-calling.md`)
-**Pattern**: `fdp-documentations/guidelines/development/ai-function-calling.md`
-**Also in this commit**: FR-048 settings-schema reservation (`FR-048-per-model-agnostic-settings.md`)
-
-## What
-
-Adds the SDK-free, vendor-neutral vocabulary for LLM function calling to `@futdevpro/fsm-dynamo/ai`,
-plus a per-model capability registry. After this, the execution layer (`nts-dynamo`, FR-047 P2+P3)
-can speak one tool shape across providers.
-
-## Why
-
-Function calling existed ONLY in the legacy OpenAI-only `FDPNTS_GPT_ControlService.getAnswerWithTools`
-(nts-fdp-templates). The strategic provider-agnostic layer had ZERO tool support: `DyFM_AI_Message`
-was `{role,content}`, `DyFM_AI_LLM_Response` had no `toolCalls`, and the `functionCalling` provider-
-capability flag was dead. This commit is the foundation for moving tool use into the strategic layer.
-
-## Design
-
-### Tool models (`src/_modules/ai/_models/`)
-```ts
-DyFM_AI_Tool          // { name, description, parameters: object /* JSON Schema */ }
-DyFM_AI_ToolCall      // { id, name, arguments: Record<string,unknown> } (normalized; OpenAI JSON-string parsed at boundary)
-DyFM_AI_ToolResult    // { toolCallId, content: string, isError? }
-DyFM_AI_ToolHandler / DyFM_AI_ToolHandlers   // lifted from legacy FDPNTS_GPT_ToolHandler(s)
-```
-
-### Additive extensions (non-breaking, optional fields)
-- `DyFM_AI_Message` + `toolCalls?` (assistant tool-call turn) + `toolCallId?` (tool-result message)
-- `DyFM_AI_LLM_Response` + `toolCalls?` + `stopReason?`
-- `DyFM_AI_MessageRole.tool`
-- `DyFM_AI_ModelType` + `Image`, `Audio`
-
-### Per-model registry
-- `DyFM_AI_ModelInfo` (id / provider / modelType / capabilities / contextWindow / maxOutputTokens
-  + lifecycle: status / successorId / sunsetDate + displayName / tier)
-- `DyFM_AI_ModelCapabilities` (functionCalling / vision / streaming / jsonMode / audioInput /
-  audioOutput / imageGeneration / embeddings)
-- Per-provider const arrays: `DyFM_OAI_Models`, `DyFM_AAI_Models`, `DyFM_GAI_Models`, `DyFM_LAI_Models`,
-  `DyFM_FdpAI_Models` (2026-06 verified data — see `[[reference_ai_provider_model_registry_research]]`)
-- `DyFM_AI_ModelRegistry_Util.findModelInfo(models, id)` / `modelSupportsTools(models, id)` — the
-  registry array is **passed in**, so the agnostic `./ai` core imports NO provider submodule
-  (keeps it SDK-free). Unknown id → `false` (honesty).
-
-### FR-048 reservation (space left, NOT implemented)
-- `DyFM_AI_ModelSettingsSchema` + `DyFM_AI_SettingConstraint` (reserved placeholder)
-- `DyFM_AI_ModelInfo.settingsSchema?` + thorough doc on `DyFM_AI_CallSettings` — the per-model
-  "which params does this model accept/reject" system. Sibling of capabilities. See `ai-per-model-settings.md`.
-
-## Spec coverage
-
-`src/_modules/ai/_collections/ai-model-registry.util.spec.ts` — findModelInfo (known/unknown) +
-modelSupportsTools (tool-capable chat / embedding=false / unknown=false). `pnpm test`: **1302/0** (70 pending).
-
-## Optional-SDK discipline
-
-The agnostic `./ai` index imports no provider submodule → `import '@futdevpro/fsm-dynamo/ai'` pulls
-zero AI SDKs. Provider model data lives under `./ai/<provider>` subpaths.
-
-## What's NOT in this commit
-
-- The agent-loop / execution (nts-dynamo P2+P3 — `requestWithTools`).
-- `adultContent` capability (added later, `de1da9b`, v01.15.18).
-- `./ai/google` + `./ai/local` subpath export-bug fix (FR-047 P5, `6a4ce9c`).
-- FR-048 runtime reconciler (reserved only).
-
-## Cross-references
-
-- `fdp-documentations/guidelines/development/ai-function-calling.md` (pattern)
-- `fdp-documentations/guidelines/development/ai-per-model-settings.md` (FR-048)
-- `NPM-packages/dynamo-nts/__documentations/2026-06-01-fr047-p2p3-function-calling.md` (the consumer)
-- legacy `NPM-packages/fdp-templates-nts/src/_modules/gpt/_services/gpt-handler.control-service.ts`

package/__documentations/2026-06-05-secret-mask-url-cred-any-scheme.md

@@ -1,37 +0,0 @@
-# 2026-06-05 — DyFM_SecretMask: url-cred pattern BARMELY scheme-re (+ elso spec)
-
-## Problema
-
-A `DyFM_SECRET_PATTERNS` `url-cred` pattern-je CSAK `http(s)://user:pass@`-t fogott
-(`/\b(https?:\/\/)([^/\s:@]+):([^\s@]+)@/g`). A `mongodb://user:pass@host` / `postgres://` /
-`redis://` connection-string jelszo — a leggyakoribb DB-secret-leak vektor a CI-logokban, es
-az FDP-stack MongoDB-t hasznal — NEM maszkolodott (runtime-igazolt gap a `fdp pipeline` secret-
-maszkolas bekotesekor).
-
-## Fix
-
-- A scheme kiszelesitve BARMELY URL-protokollra: `https?` → `[a-z][a-z0-9+.-]*` + `i` flag.
-- A user-resz `+` → `*` (ures user is): a `redis://:password@host` (username nelkuli, csak jelszo)
-  valos redis/DB auth-format.
-- A `user:pass@` szerkezet KOTELEZO marad → nincs false-positive: sima URL (`https://h:8080/path`,
-  port-tal, de `@` nelkul) NEM matchel; a `mongodb://db:27017/app` (nincs cred) NEM matchel.
-- A `$1` (scheme) + host megorzodik: `mongodb://[redacted-url-cred]@db:27017/app`.
-
-Uj regex: `/\b([a-z][a-z0-9+.-]*:\/\/)([^/\s:@]*):([^\s@]+)@/gi`
-
-## Test (uj — eddig NEM volt spec a maskerhez)
-
-Uj `secret-mask.util.spec.ts` (16 assert):
-- url-cred: http/https/mongodb/postgres/redis + case-insensitive + ures-user;
-  no-false-positive (port / cred-less URL);
-- token-regresszio: github-pat + env-style-secret;
-- non-secret: commit-SHA + verzio erintetlen;
-- `DyFM_maskSecretsDeep`: rekurziv string-maszkolas + strukturа-megorzes.
-
-`npx tsc` (exit 0) + `npx jasmine`: **1443 specs, 0 failures**, 72 pending.
-
-## Consumer
-
-Az `fdp-cli` (`pipeline-notify` Discord + `pipeline-report` Overseer-POST) mar hasznalja a
-`DyFM_maskSecrets`/`DyFM_maskSecretsDeep`-et; a broadened url-cred a kovetkezo fsm-dynamo
-dep-bump utan lep eletbe a fleeten (nem fleet-bump-olunk kezzel).