@intrect/openswarm 0.4.4 → 0.8.0

package/README.md

@@ -3,33 +3,28 @@
 [![npm version](https://img.shields.io/npm/v/@intrect/openswarm.svg)](https://www.npmjs.com/package/@intrect/openswarm)
 [![npm downloads](https://img.shields.io/npm/dm/@intrect/openswarm.svg)](https://www.npmjs.com/package/@intrect/openswarm)
 [![license](https://img.shields.io/npm/l/@intrect/openswarm.svg)](LICENSE)
-[![ko-fi](https://img.shields.io/badge/Ko--fi-support-FF5E5B?logo=ko-fi&logoColor=white)](https://ko-fi.com/unohee)
+[![SWE-bench Lite](https://img.shields.io/badge/SWE--bench_Lite-hybrid_3%2F3_resolved-2ea44f)](benchmarks/RUBRIC.md)
+[![GitHub Discussions](https://img.shields.io/github/discussions/unohee/OpenSwarm?logo=github&label=discussions)](https://github.com/unohee/OpenSwarm/discussions)
 
-> Autonomous AI agent orchestrator — Claude, GPT, Codex, and local models (Ollama/LMStudio/llama.cpp)
+> Autonomous AI agent orchestrator — Claude, GPT, Codex, **OpenRouter (any model)**, and local models (Ollama/LMStudio/llama.cpp)
 
-## Sponsor
-
-OpenSwarm is developed and maintained in my spare time by a single author.
-If the project saves you time or money, please consider chipping in —
-it directly funds ongoing updates, bug fixes, and new adapters.
-
-[![Support on Ko-fi](https://img.shields.io/badge/Support%20on-Ko--fi-FF5E5B?logo=ko-fi&logoColor=white&style=for-the-badge)](https://ko-fi.com/unohee)
-
-One-off contributions are perfectly fine — there is no subscription
-tier and no feature is paywalled. Thank you.
+> 💬 **Help shape OpenSwarm.** Share feature ideas, vote on the roadmap, and ask questions in [**GitHub Discussions**](https://github.com/unohee/OpenSwarm/discussions). The roadmap is built in the open — your feedback decides what ships next.
 
 ---
 
-OpenSwarm orchestrates multiple AI agents as autonomous code workers. It picks up Linear issues, runs Worker/Reviewer pair pipelines, reports to Discord, and retains long-term memory via LanceDB. Supports Claude Code, OpenAI GPT, Codex, and **local open-source models** via Ollama, LMStudio, or llama.cpp.
+OpenSwarm orchestrates multiple AI agents as autonomous code workers. It picks up Linear issues, runs Worker/Reviewer pair pipelines, reports to Discord, and retains long-term memory via LanceDB. Workers run on Claude Code, OpenAI GPT, Codex, **any OpenRouter model**, or **local open-source models** (Ollama, LMStudio, llama.cpp) — with cost-aware routing measured on an L0–L6 benchmark ladder.
+
+**Verified on real GitHub issues**: the agentic harness solves SWE-bench Lite instances graded by the official harness. Hybrid mode — a frontier model diagnoses read-only, a lightweight model implements with a verification loop — resolved **3/3 attempted instances** that every single lightweight model had failed, at a fraction of frontier-only cost. Workers also **learn each repository over time**: task outcomes are stored as per-repo knowledge and recalled into future prompts. ([benchmark rubric & results](benchmarks/RUBRIC.md))
 
 ## Quick Start
 
 ```bash
 npm install -g @intrect/openswarm
-openswarm
+openswarm auth login   # one-time provider setup — ChatGPT (codex/gpt) OAuth
+openswarm              # launches the TUI chat
 ```
 
-That's it. `openswarm` with no arguments launches the TUI chat interface immediately.
+`openswarm` with no arguments launches the TUI chat. You need **one provider** first: `openswarm auth login` (ChatGPT OAuth, used by `codex`/`gpt`), or `openswarm auth login --provider openrouter` / `export OPENROUTER_API_KEY=…`, or just have an authenticated `claude` on PATH. Check what's wired with `openswarm auth status`.
 
 ![TUI Chat Interface](screenshots/tui.png)
 
@@ -56,7 +51,7 @@ openswarm chat [session]         # Simple readline chat
 openswarm start                  # Start full daemon (requires config.yaml)
 openswarm run "Fix the bug" -p ~/my-project   # Run a single task
 openswarm exec "Run tests" --local --pipeline # Execute via daemon
-openswarm init                   # Generate config.yaml scaffold
+openswarm init                   # Interactive setup wizard (provider, task backend, config)
 openswarm validate               # Validate config.yaml
 
 # Code Registry & BS Detector
@@ -93,11 +88,13 @@ For autonomous operation (Linear issue processing, Discord control, PR auto-impr
 ### Prerequisites
 
 - **Node.js** >= 22
-- **Claude Code CLI** authenticated (`claude -p`) — default provider
-- **OpenAI Codex CLI** (`codex exec`) — optional alternative provider
-- **Discord Bot** token with message content intent
-- **Linear** API key and team ID
-- **GitHub CLI** (`gh`) for CI monitoring (optional)
+- **At least one LLM provider** (default `codex`):
+  - **OpenAI Codex** — default. `openswarm auth login` (ChatGPT OAuth) or a `codex` binary on PATH
+  - **OpenRouter** — `OPENROUTER_API_KEY`, or `openswarm auth login --provider openrouter`
+  - **Claude Code CLI** (`claude -p`) — opt-in fallback; an authenticated `claude` on PATH (used when codex hits its usage limit or OpenRouter is unavailable)
+  - **Local** (Ollama / LM Studio / llama.cpp) — auto-detected, no auth
+- **Native build toolchain** — `better-sqlite3` and `@lancedb/lancedb` are native modules. Prebuilt binaries cover common platforms; if yours lacks one, `npm install` builds from source and needs `python3` + a C/C++ toolchain (`build-essential` on Linux, Xcode Command Line Tools on macOS)
+- **For autonomous mode only** (optional): **Linear** API key + team ID, **Discord** bot token (message content intent), **GitHub CLI** (`gh`) for CI monitoring
 
 ### Configuration
 
@@ -133,19 +130,23 @@ LINEAR_TEAM_ID=your-linear-team-id
 ### CLI Adapter (Provider)
 
 ```yaml
-adapter: claude   # "claude" | "codex" | "gpt" | "local"
+adapter: codex   # default. "codex" | "openrouter" | "gpt" | "local" | "lmstudio" | "claude"
 ```
 
 Switch at runtime via Discord: `!provider codex` / `!provider claude`
 
 | Adapter | Backend | Models | Auth |
 |---------|---------|--------|------|
-| `claude` | Claude Code CLI | sonnet-4, haiku-4.5, opus-4 | CLI auth |
-| `codex` | OpenAI Codex CLI | o3, o4-mini | CLI auth |
+| `codex` | OpenAI Codex CLI | o3, o4-mini | OAuth PKCE / CLI auth |
+| `claude` | Claude Code CLI (`claude -p`) — opt-in fallback | sonnet-4, haiku-4.5, opus-4 | CLI auth |
 | `gpt` | OpenAI API | gpt-4o, o3, gpt-4.1 | OAuth PKCE |
+| `openrouter` | OpenRouter API (native agentic loop) | any OpenRouter model — gpt-5, gemini-2.5-flash, deepseek, glm, qwen, … | OAuth PKCE or `OPENROUTER_API_KEY` |
 | `local` | Ollama / LMStudio / llama.cpp | gemma4, llama3, mistral, qwen, etc. | None |
+| `lmstudio` | LM Studio OpenAI-compatible API | loaded LM Studio model (`LMSTUDIO_MODEL`) | Optional API key |
+
+The `openrouter` adapter runs OpenSwarm's own agentic tool loop (read/search/edit/bash with verification guards), enables ZDR (`data_collection: deny`) for non-OpenAI models, and applies Anthropic prompt caching automatically.
 
-Local models are auto-detected on standard ports (Ollama `:11434`, LMStudio `:1234`, llama.cpp `:8080`).
+Local models are auto-detected on standard ports (Ollama `:11434`, LMStudio `:1234`, llama.cpp `:8080`). Use `lmstudio` for a dedicated LM Studio endpoint (`LMSTUDIO_BASE_URL`, default `http://localhost:1234`).
 
 Per-role adapter overrides:
 
@@ -250,13 +251,15 @@ docker compose up -d         # Docker
 
 ## Features
 
-- **Multi-Provider Adapters** — Pluggable adapter system: **Claude Code**, **OpenAI GPT/Codex**, and **local models** (Ollama, LMStudio, llama.cpp) with runtime provider switching
+- **Multi-Provider Adapters** — Pluggable adapter system: **Claude Code**, **OpenAI GPT/Codex**, **OpenRouter** (any model, native agentic loop), and **local models** (Ollama, LMStudio, llama.cpp) with runtime provider switching
 - **Code Registry** — SQLite-backed entity registry tracking every function/class/type across 8 languages, with complexity scoring, test mapping, and risk assessment
 - **BS Detector** — Built-in static analysis engine that detects bad code patterns (empty catch, hardcoded secrets, `as any`, etc.) with pipeline guard integration
 - **Autonomous Pipeline** — Cron-driven heartbeat fetches Linear issues, runs Worker/Reviewer pair loops, and updates issue state automatically
 - **Worker/Reviewer Pairs** — Multi-iteration code generation with automated review, testing, and documentation stages
 - **Decision Engine** — Scope validation, rate limiting, priority-based task selection, and workflow mapping
 - **Cognitive Memory** — LanceDB vector store with Xenova/multilingual-e5-base embeddings for long-term recall across sessions
+- **Repo Knowledge Loop** — workers learn each repository over time: task outcomes (success patterns, review-rejection pitfalls) are stored per-repo and recalled into the next worker prompt
+- **SWE-bench Verified** — the agentic harness solves real SWE-bench Lite issues, graded by the official harness; hybrid mode (frontier diagnosis + lightweight implementer) resolved 3/3 attempted instances ([benchmarks/RUBRIC.md](benchmarks/RUBRIC.md))
 - **Knowledge Graph** — Static code analysis, dependency mapping, impact analysis, and file-level conflict detection across concurrent tasks
 - **Discord Control** — Full command interface for monitoring, task dispatch, scheduling, provider switching, and pair session management
 - **Rich TUI Chat** — Claude Code inspired terminal interface with tabs, streaming responses, and geek-themed loading messages
@@ -294,6 +297,23 @@ Memory types: `belief` · `strategy` · `user_model` · `system_pattern` · `con
 
 Background: decay, consolidation, contradiction detection, distillation.
 
+**Repo knowledge loop** — every completed task writes repo-scoped knowledge
+(success → `system_pattern` with files changed + approach, review rejection →
+`constraint` pitfall), and the next task on the same repo recalls the most
+relevant entries into the worker prompt as a "Repository Knowledge" section.
+Workers get better at a codebase the more they work on it.
+
+### Benchmarks (L0–L6)
+
+`benchmarks/` contains a difficulty ladder for routing models by measured
+capability — synthetic L0–L5 tasks with deterministic grading, and L6 = real
+GitHub issues (SWE-bench Lite) solved by the OpenSwarm harness and graded by
+the official swebench harness. Headline: **hybrid mode** (frontier read-only
+diagnosis + lightweight implementer with a verification loop) resolved 3/3
+attempted instances that every single lightweight model had failed. See
+[benchmarks/RUBRIC.md](benchmarks/RUBRIC.md) for the rubric, measured results,
+and the harness defects the benchmark uncovered.
+
 ---
 
 ## Discord Commands

package/config.example.yaml

@@ -3,9 +3,15 @@
 # Copy this file to config.yaml to use
 
 # Default CLI adapter for worker/reviewer stages
-# Options: claude, codex, gpt, local
-# For GPT: run `openswarm auth login --provider gpt` first
+# Options: claude, codex, gpt, local, lmstudio, openrouter
+# For GPT: run `openswarm auth login --provider gpt`
+#   (uses the public Codex OAuth client by default — no extra config needed)
+# For OpenRouter: run `openswarm auth login --provider openrouter`
+#   (PKCE browser flow → stores a sk-or-* API key; falls back to manual paste)
 # For local: start Ollama, LMStudio, or llama.cpp server
+# For lmstudio: start LM Studio Local Server (default http://localhost:1234)
+#   Optional env: LMSTUDIO_BASE_URL, LMSTUDIO_MODEL, LMSTUDIO_API_KEY
+#   If LMSTUDIO_MODEL is unset, the adapter auto-selects the first loaded model.
 adapter: claude
 
 discord:
@@ -49,7 +55,7 @@ autonomous:
   decomposition:
     enabled: true                    # Enable decomposition
     thresholdMinutes: 30             # Decompose if estimated time exceeds this
-    plannerModel: claude-sonnet-4-20250514  # Planner model
+    plannerModel: claude-opus-4-7    # Planner model (Opus for deep decomposition)
 
   # Per-role settings
   # Hybrid config: Claude for complex coding, local model for review/docs ($0)
@@ -57,15 +63,15 @@ autonomous:
     worker:
       enabled: true
       adapter: claude
-      model: claude-sonnet-4-20250514         # Sonnet for coding tasks
-      escalateModel: claude-opus-4-6          # On failure: Opus
+      model: claude-sonnet-4-6         # Sonnet for coding tasks
+      escalateModel: claude-opus-4-7          # On failure: Opus
       escalateAfterIteration: 3
       timeoutMs: 1800000  # 30 minutes
     reviewer:
       enabled: true
       adapter: local                          # Local model — free, 7s response
       model: gemma-4-e4b-it                   # Gemma 4 e4b via LMStudio
-      escalateModel: claude-sonnet-4-20250514 # Spot check: Sonnet reviews after N revisions
+      escalateModel: claude-sonnet-4-6 # Spot check: Sonnet reviews after N revisions
       escalateAfterIteration: 3               # Escalate from 3rd iteration
       timeoutMs: 60000    # 1 minute (local models are slower)
     tester:
@@ -88,13 +94,19 @@ autonomous:
 
   # Pipeline guards
   guards:
-    qualityGate: true
+    qualityGate: true       # bad-edit lint gate: tsc/ruff on changed files (blocking)
     fakeDataGuard: true
     conventionalCommits: true
     branchValidation: true
     uncertaintyDetection: true
     registryCheck: true
-    bsDetector: true
+    bsDetector: true        # blocks on critical code-smell patterns
+
+  # Self-repair reflection budget: max objective (lint/bs/test) failures tolerated
+  # before the loop gives up on bad edits. Lower it to cap token burn when
+  # reflection stops making progress; the loop also bails early on stagnation
+  # (identical errors twice in a row). Default: 3.
+  maxReflections: 3
 
   # Job profiles for lightweight vs heavy work
   jobProfiles:
@@ -105,8 +117,8 @@ autonomous:
     - name: deep-engineering
       minMinutes: 16
       roles:
-        worker: claude-sonnet-4-20250514      # Sonnet for complex work
-        reviewer: claude-sonnet-4-20250514    # Sonnet reviews Sonnet
+        worker: claude-sonnet-4-6      # Sonnet for complex work
+        reviewer: claude-sonnet-4-6    # Sonnet reviews Sonnet
 
 # Long-running task monitoring (RunPod training, batch processing, etc.)
 #

package/dist/adapters/agenticLoop.d.ts

@@ -1,4 +1,4 @@
-import { type ToolDefinition } from './tools.js';
+import { type ToolCall, type ToolDefinition } from './tools.js';
 import type { CliRunResult } from './types.js';
 /** OpenAI Chat Completions API 메시지 포맷 */
 export type ChatMessage = {
@@ -51,7 +51,7 @@ export interface AgenticLoopOptions {
     model: string;
     /** API 호출 함수 (어댑터별로 주입) */
     callApi: (messages: ChatMessage[], tools: ToolDefinition[]) => Promise<ChatCompletionResponse>;
-    /** 최대 도구 사용 턴 수 (기본: 15) */
+    /** 최대 도구 사용 턴 수 (기본: 20) */
     maxTurns?: number;
     /** 전체 타임아웃 (ms, 기본: 300000) */
     timeoutMs?: number;
@@ -59,6 +59,29 @@ export interface AgenticLoopOptions {
     onLog?: (line: string) => void;
     /** 도구 사용 허용 여부 (기본: true) */
     enableTools?: boolean;
+    /** 토큰 기반 압축 트리거 임계값 (기본: 24000) */
+    compactTokenThreshold?: number;
+    /** 이 메시지 수를 넘어야 압축 후보 (VEGA compact_threshold, 기본: 24) */
+    compactAfterMessages?: number;
+    /** 압축 시 항상 원본 유지할 최근 메시지 수 (VEGA keep_recent, 기본: 8) */
+    keepRecentMessages?: number;
+    /**
+     * 수정이 필수인 작업의 no-edit 종료 가드. 모델이 edit/write 도구를 한 번도 안 쓰고
+     * 최종 텍스트로 끝내려 하면 "아직 수정 안 했다, 계속하라"고 N회까지 되민다.
+     * 경량 모델(gemini 등)이 탐색만 하고 일찍 결론 내는 패턴 차단 (SWE 하이브리드에서 발견).
+     * 기본 0 (비활성) — 수정 없는 작업(진단·분석)도 정상이므로 옵트인.
+     */
+    nudgeMaxOnNoEdit?: number;
+    /** Verification-harness files for which edit/write are refused (see tools.ts ToolExecOptions) */
+    protectedFiles?: string[];
+    /** bash tool timeout — docker-based tests need minutes (default 30s) */
+    bashTimeoutMs?: number;
+    /** Expose web_fetch + web_search tools (default true). Disabled e.g. for SWE-bench integrity. */
+    webTools?: boolean;
+    /** MCP tools (named `server__tool`) discovered from mcp.json, exposed alongside the native tools. */
+    mcpTools?: ToolDefinition[];
+    /** Abort the loop (checked each turn) — Esc/Ctrl+C in chat. */
+    signal?: AbortSignal;
 }
 /** 루프 실행 결과 */
 export interface AgenticLoopResult {
@@ -86,5 +109,25 @@ export declare function runAgenticLoop(options: AgenticLoopOptions): Promise<Age
  * AgenticLoopResult → CliRunResult 변환
  */
 export declare function loopResultToCliResult(result: AgenticLoopResult): CliRunResult;
+/** Stable key for a tool call (name + args) — used to detect repeated calls. */
+export declare function toolCallKey(tc: ToolCall): string;
+/**
+ * True when every tool call this turn was already seen (same name+args), i.e.
+ * pure repetition with no new info or change — a stalled turn. Empty turns are
+ * not stalls (the model produced no tool calls, which ends the loop normally).
+ */
+export declare function allToolCallsSeen(toolCalls: ToolCall[], seen: Set<string>): boolean;
+/**
+ * 이전 턴(assistant+tool 쌍)을 요약 1줄로 교체.
+ * OpenAI API 제약: tool 메시지는 직전 assistant의 tool_call_id와 대응해야 하므로
+ * 오래된 assistant+tool 쌍은 텍스트 요약으로 대체해 API 오류를 방지.
+ *
+ * 보존 기준 (VEGA keep_recent): 최근 keepRecent개 메시지 블록은 항상 원본 유지.
+ * tool 메시지는 직전 assistant의 tool_call_id와 짝이 맞아야 하므로, 보존 경계는
+ * keepRecent 지점 이후 첫 assistant로 정렬해 짝이 깨진 tool 메시지가 남지 않게 한다.
+ * 기존 [Prior turns compacted] 요약이 있으면 새 요약에 합산 후 교체.
+ * (테스트를 위해 export — 외부에서 직접 호출할 일은 없음)
+ */
+export declare function compactPriorTurns(messages: ChatMessage[], keepRecent?: number): void;
 export {};
 //# sourceMappingURL=agenticLoop.d.ts.map

package/dist/adapters/agenticLoop.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"agenticLoop.d.ts","sourceRoot":"","sources":["../../src/adapters/agenticLoop.ts"],"names":[],"mappings":"AAQA,OAAO,EAAsE,KAAK,cAAc,EAAE,MAAM,YAAY,CAAC;AACrH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAI/C,yCAAyC;AACzC,MAAM,MAAM,WAAW,GACnB;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACnC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACjC;IAAE,IAAI,EAAE,WAAW,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,UAAU,CAAC,EAAE,WAAW,EAAE,CAAA;CAAE,GACzE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAE5D,UAAU,WAAW;IACnB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,UAAU,sBAAsB;IAC9B,OAAO,EAAE,KAAK,CAAC;QACb,OAAO,EAAE;YACP,IAAI,EAAE,MAAM,CAAC;YACb,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;YACvB,UAAU,CAAC,EAAE,WAAW,EAAE,CAAC;SAC5B,CAAC;QACF,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC,CAAC;IACH,KAAK,CAAC,EAAE;QACN,aAAa,EAAE,MAAM,CAAC;QACtB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED,iBAAiB;AACjB,MAAM,WAAW,kBAAkB;IACjC,eAAe;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,uBAAuB;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,+BAA+B;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU;IACV,KAAK,EAAE,MAAM,CAAC;IACd,2BAA2B;IAC3B,OAAO,EAAE,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,KAAK,EAAE,cAAc,EAAE,KAAK,OAAO,CAAC,sBAAsB,CAAC,CAAC;IAC/F,4BAA4B;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,+BAA+B;IAC/B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,gBAAgB;IAChB,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC/B,6BAA6B;IAC7B,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,eAAe;AACf,MAAM,WAAW,iBAAiB;IAChC,gBAAgB;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,mBAAmB;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,kBAAkB;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAID;;;;;;;GAOG;AACH,wBAAsB,cAAc,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,iBAAiB,CAAC,CA0H5F;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,iBAAiB,GAAG,YAAY,CAO7E"}
+{"version":3,"file":"agenticLoop.d.ts","sourceRoot":"","sources":["../../src/adapters/agenticLoop.ts"],"names":[],"mappings":"AAQA,OAAO,EAAuD,KAAK,QAAQ,EAAmB,KAAK,cAAc,EAAE,MAAM,YAAY,CAAC;AAEtI,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAuC/C,yCAAyC;AACzC,MAAM,MAAM,WAAW,GACnB;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACnC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACjC;IAAE,IAAI,EAAE,WAAW,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,UAAU,CAAC,EAAE,WAAW,EAAE,CAAA;CAAE,GACzE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAE5D,UAAU,WAAW;IACnB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,UAAU,sBAAsB;IAC9B,OAAO,EAAE,KAAK,CAAC;QACb,OAAO,EAAE;YACP,IAAI,EAAE,MAAM,CAAC;YACb,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;YACvB,UAAU,CAAC,EAAE,WAAW,EAAE,CAAC;SAC5B,CAAC;QACF,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC,CAAC;IACH,KAAK,CAAC,EAAE;QACN,aAAa,EAAE,MAAM,CAAC;QACtB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED,iBAAiB;AACjB,MAAM,WAAW,kBAAkB;IACjC,eAAe;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,uBAAuB;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,+BAA+B;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU;IACV,KAAK,EAAE,MAAM,CAAC;IACd,2BAA2B;IAC3B,OAAO,EAAE,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,KAAK,EAAE,cAAc,EAAE,KAAK,OAAO,CAAC,sBAAsB,CAAC,CAAC;IAC/F,4BAA4B;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,+BAA+B;IAC/B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,gBAAgB;IAChB,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;IAC/B,6BAA6B;IAC7B,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,mCAAmC;IACnC,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,0DAA0D;IAC1D,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,wDAAwD;IACxD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,iGAAiG;IACjG,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,wEAAwE;IACxE,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,iGAAiG;IACjG,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,qGAAqG;IACrG,QAAQ,CAAC,EAAE,cAAc,EAAE,CAAC;IAC5B,+DAA+D;IAC/D,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,eAAe;AACf,MAAM,WAAW,iBAAiB;IAChC,gBAAgB;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,mBAAmB;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,kBAAkB;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAID;;;;;;;GAOG;AACH,wBAAsB,cAAc,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAkP5F;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,iBAAiB,GAAG,YAAY,CAO7E;AAED,gFAAgF;AAChF,wBAAgB,WAAW,CAAC,EAAE,EAAE,QAAQ,GAAG,MAAM,CAEhD;AAED;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,SAAS,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,OAAO,CAGlF;AAID;;;;;;;;;;GAUG;AACH,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,UAAU,SAAI,GAAG,IAAI,CAwD/E"}