fixo-cli 1.0.4 → 2.0.0

package/CHANGELOG.md

@@ -0,0 +1,62 @@
+# Changelog
+
+All notable changes to FixO CLI will be documented in this file.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and this project adheres to [Semantic Versioning](https://semver.org/).
+
+---
+
+## [1.0.4] – 2025-06-26
+
+### Security
+- **decryptKey** now throws on AES-256-GCM decryption failure instead of silently returning ciphertext, preventing corrupted keys from being used as live credentials.
+- `getOrCreateRunId()` switched from `Math.random()` to `crypto.randomBytes(6)` for cryptographically secure staging-directory namespace IDs.
+- `RETRYABLE_STATUS_CODES` in `agent-client.ts` now includes `504` (Gateway Timeout), matching the canonical set in `retry.ts`.
+
+### Bug Fixes
+- Fixed a duplicate `name === 'AbortError'` condition in `defaultIsRetryable` (dead-code bug in `retry.ts`).
+- `SIGINT` handler is now deduplicated when both the readline interface and the process fire simultaneously.
+- `buildLavaStatusState()` now derives the `transport` field from the actual `provider_mode` config instead of always displaying `'freellmapi'`.
+- `getOrCreateRunId()` uses canonical `MUTATION_TOOL_NAMES` set instead of a fragile string-heuristic for mutating action detection.
+
+### Improvements
+- Non-null assertions (`!`) in setup-wizard provider registry lookups replaced with proper runtime guards.
+- Removed dead empty section headers from `src/ui/prompt.ts`.
+- Simplified `buildLavaStatusState()` ternary chain (removed unreachable `else` branch).
+- Removed unused `width` variable from `drawSuggestions()`.
+- Silent `catch {}` in `exitCleanup` now logs in debug/verbose mode.
+- Trailing whitespace removed from `retry.ts`.
+
+### Packaging
+- Added `"exports"` field to `package.json` for proper ESM resolution.
+- Added `postinstall` script to enforce Node.js >= 20.0.0 at install time.
+- `CHANGELOG.md` added to published `files` list.
+
+---
+
+## [1.0.3] – 2025-06-20
+
+### Added
+- Atomic staging pipeline with rollback (`AtomicStagingManager`).
+- LSP pre-save gate (Pillar 3) for syntax validation before disk writes.
+- Semantic loop detector (`SemanticLoopDetector`) to complement hash-based loop trap.
+- `run_command_async` / `poll_command_status` / `kill_command` tools for long-running tasks.
+- `glob_files` tool using Node.js 22+ native `fs.promises.glob`.
+
+### Security
+- AES-256-GCM encryption for API keys at rest in `providers.json`.
+- `WorkspaceGuard.assertNotPlatformPath()` prevents agent from modifying its own source files.
+- `SCRUB_PATTERNS` expanded to cover OpenAI, Anthropic, OpenRouter, GitHub, Google, AWS, and JWT tokens.
+- Provider credential vault (`ProviderKeyVault`) with scoped `withApiKey` callbacks.
+
+---
+
+## [1.0.0] – 2025-05-01
+
+### Added
+- Initial release of FixO CLI.
+- Multi-provider support: OpenAI, Anthropic, Groq, Google, Mistral, Together, Perplexity, DeepSeek, Cohere, OpenRouter, NVIDIA, xAI, GitHub Models, Ollama, Zen.
+- FreeLLMAPI proxy mode with load-balanced failover.
+- Interactive setup wizard (`/setup`).
+- Loop-trap detection, atomic writes, LSP integration.
+- REPL with slash commands, autocomplete, paste attachments, and session history.

package/README.md

@@ -4,11 +4,11 @@
 [![Language](https://img.shields.io/badge/Language-TypeScript-blue.svg)](https://www.typescriptlang.org/)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 [![Engine](https://img.shields.io/badge/Engine-Tree--Sitter-orange.svg)](https://tree-sitter.github.io/tree-sitter/)
-[![Status](https://img.shields.io/badge/Status-Production--Ready-brightgreen.svg)]()
+[![Status](https://img.shields.io/badge/Status-Beta-yellow.svg)]()
 
-Fixo CLI is a terminal-based autonomous coding assistant designed to execute complex programming tasks directly in your workspace. Built as a self-correcting agent, it analyzes code using abstract syntax trees (AST), writes implementation plans, edits code files, runs test suites, and iterates until the goal is fully achieved.
+Fixo CLI is a terminal-based autonomous coding assistant designed to execute complex programming tasks directly in your workspace. It writes implementation plans, edits code files, runs test suites, and iterates toward the goal. Tree-sitter is used today for shell-command parsing and LSP-fallback syntax checks; expanding it to the workspace symbol map is on the roadmap.
 
-Fixo CLI integrates seamlessly with **FreeLLMAPI**, automatically load-balancing and failing over across **20+ free LLM providers** (such as Gemini, Groq, SambaNova, Cerebras, and NVIDIA NIM) for zero-cost, state-of-the-art agentic coding.
+Fixo CLI ships with **13 direct providers built-in** (OpenAI, Anthropic, Google, Groq, Mistral, Cohere, OpenRouter, NVIDIA, Cerebras, SambaNova, GitHub Models, xAI, Zen) — paste your own key for any of them. The optional **FreeLLMAPI** proxy backend is available as an opt-in convenience for users who want load-balanced failover across free-tier providers without managing individual keys.
 
 ---
 
@@ -18,33 +18,35 @@ Here is how Fixo CLI compares against other prominent terminal and editor-based
 
 | Feature / Metric | **Fixo CLI** | **Claude Code** | **Aider** | **Cline** |
 | :--- | :--- | :--- | :--- | :--- |
-| **API Cost** | 💰 **100% Free** (via FreeLLMAPI) | 💸 **Paid** (Anthropic API charges) | 💸 **Paid** (Requires personal keys) | 💸 **Paid** (Requires personal keys) |
-| **Multi-Provider Fallback**| 🔄 **Automatic Failover** (No interruptions) | ❌ None (Locked to Anthropic) | ❌ Manual (Requires editing configs) | ❌ Manual (Drops request on 429) |
-| **Workspace Indexing** | 🌳 **AST / Tree-Sitter** (Semantic map) | 🔍 Regex / basic grep | 🗺️ Git/ctags-based map | 🔍 Basic file search |
-| **Autonomy Loops** | 🤖 **Multi-agent / Planning Mode** | 🤖 Agent loops | 💬 Interactive / chat-driven | 💬 Prompt-to-action loops |
-| **Self-Correction** | 🧪 **Built-in test runner & loops** | ❌ Manual trigger | ❌ Requires manual input | ❌ Requires manual input |
-| **No-Card Verification** | ✅ **Yes** (Zero billing required) | ❌ No (Requires credit card) | ❌ No (Requires paid API keys) | ❌ No (Requires paid API keys) |
+| **API Cost** | 💰 **BYOK or free via optional FreeLLMAPI proxy** | 💸 **Paid** (Anthropic API charges) | 💸 **Paid** (Requires personal keys) | 💸 **Paid** (Requires personal keys) |
+| **Multi-Provider Fallback**| 🔄 **Automatic failover (FreeLLMAPI proxy mode)** | ❌ None (Locked to Anthropic) | ❌ Manual (Requires editing configs) | ❌ Manual (Drops request on 429) |
+| **Workspace Indexing** | 🗂️ Depth-capped regex scan (tree-sitter symbol map planned) | 🔍 Regex / basic grep | 🗺️ Git/ctags-based map | 🔍 Basic file search |
+| **Autonomy Loops** | 🤖 **Multi-agent / Planning Mode** ¹ | 🤖 Agent loops | 💬 Interactive / chat-driven | 💬 Prompt-to-action loops |
+| **Self-Correction** | 🧪 Opt-in via `/fix-tests` (automatic post-edit verification on the roadmap) | ❌ Manual trigger | ❌ Requires manual input | ❌ Requires manual input |
+| **No-Card Verification** | ✅ **Yes** (BYOK or free proxy — no card required either way) | ❌ No (Requires credit card) | ❌ No (Requires paid API keys) | ❌ No (Requires paid API keys) |
+
+¹ The multi-agent path is currently triggered by a keyword heuristic. An LLM-based complexity classifier is wired-but-dead in `src/planner.ts` and is being moved onto the live code path.
 
 ---
 
 ## ⚙️ Architecture & Lifecycle Flow
 
-Fixo CLI separates concerns between code understanding (AST parser), task coordination (Planner), and execution (Agent). 
+Fixo CLI separates concerns between code understanding (workspace indexer), task coordination (Planner), and execution (Agent). 
 
 ```mermaid
 sequenceDiagram
     autonumber
     actor User as Developer
     participant CLI as Fixo CLI
-    participant Indexer as AST Workspace Indexer
+    participant Indexer as Workspace Indexer
     participant Planner as Plan Engine
     participant Agent as Autonomous Agent
     participant Proxy as FreeLLMAPI Proxy
     participant LLM as Provider (Groq/Gemini/NIM)
 
     User->>CLI: Request task (e.g. "Fix auth bug")
-    CLI->>Indexer: Scan repository & generate AST maps
-    Indexer-->>CLI: Return semantic codebase layout
+    CLI->>Indexer: Scan repository & build export map
+    Indexer-->>CLI: Return repository structure summary
     CLI->>Planner: Propose implementation plan
     Planner->>Proxy: Fetch reasoning (smart routing)
     Proxy->>LLM: Try highest ranked provider
@@ -61,12 +63,14 @@ sequenceDiagram
     Agent-->>User: Task completed successfully!
 ```
 
+> **Note on routing:** Smart routing across models (cheaper for planning, stronger for execution) is a feature of the optional FreeLLMAPI proxy backend. In direct-provider mode, requests use the model you selected at setup — local fast/heavy-tier substitution via `preferences.modelRouting` is on the roadmap.
+
 ---
 
 ## 🌟 Key Features
 
 * **Autonomous Agent Loop:** Fixo CLI runs an agent loop that defines planning sub-agents, writes files, runs shell commands, reads compiler output, and self-corrects until tests pass.
-* **Workspace AST Indexer:** Uses **Tree-Sitter** to parse JavaScript, TypeScript, Python, and Go codebases, generating a semantic repository map for precise context insertion.
+* **Workspace Indexer:** Today, a depth-capped (4 levels, 200 files) directory scanner extracts exports via regex for a quick repository map. A tree-sitter-backed symbol map for TS/JS/Python/Go/Rust is on the roadmap; the WASM runtime is already vendored and used elsewhere in the codebase.
 * **Free Multi-Provider Routing:** Connects to your FreeLLMAPI server to query models like Llama 3.3, Qwen 3, and Gemini 2.5/3.1 without incurring high API costs.
 * **Smart Cooldown & Failover:** The CLI automatically tracks rate-limited providers (429/402/404) and switches to working alternatives in the fallback chain mid-request.
 * **Resilience Stack:** Stream recovery, provider cooldown, context-budget enforcement, and a local telemetry sink work together so the agent stays productive on flaky networks and large codebases. See [Resilience](#-resilience) below.

package/dist/agent/agent-client.d.ts

@@ -4,6 +4,7 @@
  * Includes retry with exponential backoff for transient errors.
  */
 import type { ChatMessage, ChatToolDefinition, ChatToolChoice, TokenUsage } from '../shared/types.js';
+import { type ModelRoutingConfig } from '../config.js';
 export interface ChatOptions {
     tools?: ChatToolDefinition[];
     tool_choice?: ChatToolChoice;
@@ -68,19 +69,40 @@ export declare class HttpError extends Error {
     status: number;
     constructor(status: number, message: string);
 }
+/**
+ * Thrown when the client is running in direct-provider mode but the
+ * model the caller asked for did not resolve to any direct provider
+ * via {@link AgentClient.resolveDirectConfig}. Catching this gives
+ * the UI a chance to suggest `/model` or `/providers add` instead of
+ * silently leaking the request to the FreeLLMAPI proxy.
+ */
+export declare class DirectModelUnresolvedError extends Error {
+    model: string;
+    constructor(model: string);
+}
 export declare class AgentClient {
     private baseUrl;
     private apiKey;
     private verbose;
-    constructor(apiKey: string, apiUrl?: string, verbose?: boolean);
+    private providerMode;
+    private modelRouting;
+    constructor(apiKey: string, apiUrl?: string, verbose?: boolean, providerMode?: 'direct' | 'proxy', modelRouting?: ModelRoutingConfig);
+    /**
+     * Phase 2.4 — substitute the caller-supplied model with a
+     * configured tier when `required_capabilities` asks for one.
+     * Returns the caller's model unchanged when no matching tier is
+     * configured, so the call is a no-op for users who haven't set
+     * up routing.
+     */
+    private applyCapabilityRouting;
     private resolveDirectConfig;
     /**
-     * Maps a model id to the provider that will actually serve the
-     * request — used as the key for `providerCooldown` tracking. The
-     * `freellmapi` sentinel covers the proxy path; everything else
-     * routes through a direct provider.
+     * Maps a model id to the tracking key for `providerCooldown`.
+     * Model-specific isolation ensures a timeout on one model (e.g.
+     * `openrouter:claude-3`) does not poison other models on the
+     * same provider gateway.
      */
-    private getProviderId;
+    private getCooldownKey;
     chat(messages: ChatMessage[], model: string, options?: ChatOptions): Promise<ChatResult>;
     private executeSingleChatStreamAttempt;
     chatStream(messages: ChatMessage[], model: string, options?: ChatOptions): AsyncGenerator<StreamChunk>;

package/dist/agent/agent-client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"agent-client.d.ts","sourceRoot":"","sources":["../../src/agent/agent-client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,OAAO,KAAK,EACV,WAAW,EAGX,kBAAkB,EAClB,cAAc,EACd,UAAU,EACX,MAAM,oBAAoB,CAAC;AAgD5B,MAAM,WAAW,WAAW;IAC1B,KAAK,CAAC,EAAE,kBAAkB,EAAE,CAAC;IAC7B,WAAW,CAAC,EAAE,cAAc,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,eAAe,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,UAAU,GAAG,UAAU,GAAG,UAAU,GAAG,eAAe,CAAC;IAC7F,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;IACjC;uEACmE;IACnE,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,UAAU,EAAE,KAAK,CAAC;QAChB,EAAE,EAAE,MAAM,CAAC;QACX,IAAI,EAAE,UAAU,CAAC;QACjB,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,SAAS,EAAE,MAAM,CAAA;SAAE,CAAC;KAC/C,CAAC,GAAG,IAAI,CAAC;IACV,KAAK,EAAE,UAAU,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,SAAS,GAAG,UAAU,GAAG,iBAAiB,GAAG,iBAAiB,GAAG,MAAM,CAAC;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE;QACV,KAAK,EAAE,MAAM,CAAC;QACd,EAAE,CAAC,EAAE,MAAM,CAAC;QACZ,QAAQ,CAAC,EAAE;YAAE,IAAI,CAAC,EAAE,MAAM,CAAC;YAAC,SAAS,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC;KAClD,CAAC;IACF,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC/B;AAID,oBAAY,WAAW;IACrB,IAAI,SAAS;IACb,QAAQ,aAAa;CACtB;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAa;IAC7B,OAAO,CAAC,SAAS,CAAc;IAC/B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,aAAa,CAAkB;IAEvC,IAAI,aAAa,IAAI,OAAO,CAE3B;IAEA,IAAI,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC,YAAY,CAAC;IAoB/C,OAAO,CAAC,oBAAoB;IAgD5B,OAAO,CAAC,mBAAmB;IAkC3B,KAAK,IAAI,YAAY,GAAG,IAAI;CAS7B;AAID,qBAAa,SAAU,SAAQ,KAAK;IAClC,MAAM,EAAE,MAAM,CAAC;gBACH,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM;CAK5C;AAID,qBAAa,WAAW;IACtB,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,OAAO,CAAU;gBAEb,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,EAAE,OAAO,UAAQ;IAM5D,OAAO,CAAC,mBAAmB;IAqF3B;;;;;OAKG;IACH,OAAO,CAAC,aAAa;IAQf,IAAI,CACR,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,GACxB,OAAO,CAAC,UAAU,CAAC;YAwNP,8BAA8B;IAgStC,UAAU,CACf,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,GACxB,cAAc,CAAC,WAAW,CAAC;IAkL9B;;;;;;;;;;;;;;;;;;;OAmBG;IACI,oBAAoB,CACzB,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,EACzB,iBAAiB,GAAE,MAAU,GAC5B,cAAc,CAAC,WAAW,EAAE,IAAI,EAAE,IAAI,CAAC;IA+GpC,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,SAA2B,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC;IA+E/E,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC;CAa/B"}
+{"version":3,"file":"agent-client.d.ts","sourceRoot":"","sources":["../../src/agent/agent-client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,OAAO,KAAK,EACV,WAAW,EAGX,kBAAkB,EAClB,cAAc,EACd,UAAU,EACX,MAAM,oBAAoB,CAAC;AAS5B,OAAO,EAAmB,KAAK,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAsDxE,MAAM,WAAW,WAAW;IAC1B,KAAK,CAAC,EAAE,kBAAkB,EAAE,CAAC;IAC7B,WAAW,CAAC,EAAE,cAAc,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,eAAe,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,UAAU,GAAG,UAAU,GAAG,UAAU,GAAG,eAAe,CAAC;IAC7F,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;IACjC;uEACmE;IACnE,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,UAAU,EAAE,KAAK,CAAC;QAChB,EAAE,EAAE,MAAM,CAAC;QACX,IAAI,EAAE,UAAU,CAAC;QACjB,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,SAAS,EAAE,MAAM,CAAA;SAAE,CAAC;KAC/C,CAAC,GAAG,IAAI,CAAC;IACV,KAAK,EAAE,UAAU,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,SAAS,GAAG,UAAU,GAAG,iBAAiB,GAAG,iBAAiB,GAAG,MAAM,CAAC;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE;QACV,KAAK,EAAE,MAAM,CAAC;QACd,EAAE,CAAC,EAAE,MAAM,CAAC;QACZ,QAAQ,CAAC,EAAE;YAAE,IAAI,CAAC,EAAE,MAAM,CAAC;YAAC,SAAS,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC;KAClD,CAAC;IACF,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC/B;AAID,oBAAY,WAAW;IACrB,IAAI,SAAS;IACb,QAAQ,aAAa;CACtB;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAa;IAC7B,OAAO,CAAC,SAAS,CAAc;IAC/B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,aAAa,CAAkB;IAEvC,IAAI,aAAa,IAAI,OAAO,CAE3B;IAEA,IAAI,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC,YAAY,CAAC;IAoB/C,OAAO,CAAC,oBAAoB;IAgD5B,OAAO,CAAC,mBAAmB;IAkC3B,KAAK,IAAI,YAAY,GAAG,IAAI;CAS7B;AAID,qBAAa,SAAU,SAAQ,KAAK;IAClC,MAAM,EAAE,MAAM,CAAC;gBACH,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM;CAK5C;AAID;;;;;;GAMG;AACH,qBAAa,0BAA2B,SAAQ,KAAK;IAChC,KAAK,EAAE,MAAM;gBAAb,KAAK,EAAE,MAAM;CAOjC;AAED,qBAAa,WAAW;IACtB,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,OAAO,CAAU;IACzB,OAAO,CAAC,YAAY,CAAqB;IACzC,OAAO,CAAC,YAAY,CAAqB;gBAGvC,MAAM,EAAE,MAAM,EACd,MAAM,CAAC,EAAE,MAAM,EACf,OAAO,UAAQ,EACf,YAAY,GAAE,QAAQ,GAAG,OAAiB,EAC1C,YAAY,CAAC,EAAE,kBAAkB;IASnC;;;;;;OAMG;IACH,OAAO,CAAC,sBAAsB;IAc9B,OAAO,CAAC,mBAAmB;IAqF3B;;;;;OAKG;IACH,OAAO,CAAC,cAAc;IAQhB,IAAI,CACR,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,GACxB,OAAO,CAAC,UAAU,CAAC;YAqOP,8BAA8B;IAgStC,UAAU,CACf,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,GACxB,cAAc,CAAC,WAAW,CAAC;IAwL9B;;;;;;;;;;;;;;;;;;;OAmBG;IACI,oBAAoB,CACzB,QAAQ,EAAE,WAAW,EAAE,EACvB,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,WAAgB,EACzB,iBAAiB,GAAE,MAAU,GAC5B,cAAc,CAAC,WAAW,EAAE,IAAI,EAAE,IAAI,CAAC;IA+GpC,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,SAA2B,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC;IA+E/E,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC;CAa/B"}

package/dist/agent/agent-client.js

@@ -9,8 +9,23 @@ import { extractTextFromContent } from '../shared/content.js';
 /* ──────────────────────── Constants ──────────────────────── */
 const MAX_RETRIES = 5;
 const BASE_DELAY_MS = 1500;
-const RETRYABLE_STATUS_CODES = new Set([408, 429, 500, 502, 503]);
-const BASE_URL = process.env.FIXO_API_URL || DEFAULT_API_URL;
+const RETRYABLE_STATUS_CODES = new Set([408, 429, 500, 502, 503, 504]);
+function getValidatedApiUrl(urlStr) {
+    if (!urlStr)
+        return undefined;
+    try {
+        const parsed = new URL(urlStr);
+        if (parsed.protocol === 'http:' && parsed.hostname !== 'localhost' && parsed.hostname !== '127.0.0.1') {
+            console.warn(`[Security Warning] API URL is using an insecure HTTP protocol (${urlStr}). HTTPS is required for remote URLs. Falling back to default.`);
+            return undefined;
+        }
+        return urlStr;
+    }
+    catch {
+        return undefined;
+    }
+}
+const BASE_URL = getValidatedApiUrl(process.env.FIXO_API_URL) || DEFAULT_API_URL;
 /** Wrapper around `providerCooldown.recordFailure` that also emits a
  *  telemetry event. Keeps the 6 callsites terse. */
 function trackProviderError(providerId, status, message) {
@@ -152,14 +167,55 @@ export class HttpError extends Error {
     }
 }
 /* ──────────────────────── AgentClient ──────────────────────── */
+/**
+ * Thrown when the client is running in direct-provider mode but the
+ * model the caller asked for did not resolve to any direct provider
+ * via {@link AgentClient.resolveDirectConfig}. Catching this gives
+ * the UI a chance to suggest `/model` or `/providers add` instead of
+ * silently leaking the request to the FreeLLMAPI proxy.
+ */
+export class DirectModelUnresolvedError extends Error {
+    model;
+    constructor(model) {
+        super(`Model "${model}" did not match any direct provider configured in your vault. ` +
+            `Run /providers to add a key, /model to pick a recognized model, or run setup again to switch to FreeLLMAPI proxy mode.`);
+        this.model = model;
+        this.name = 'DirectModelUnresolvedError';
+    }
+}
 export class AgentClient {
     baseUrl;
     apiKey;
     verbose;
-    constructor(apiKey, apiUrl, verbose = false) {
-        this.baseUrl = process.env.FIXO_API_URL || apiUrl || BASE_URL;
+    providerMode;
+    modelRouting;
+    constructor(apiKey, apiUrl, verbose = false, providerMode = 'proxy', modelRouting) {
+        this.baseUrl = getValidatedApiUrl(process.env.FIXO_API_URL) || getValidatedApiUrl(apiUrl) || BASE_URL;
         this.apiKey = apiKey;
         this.verbose = verbose;
+        this.providerMode = providerMode;
+        this.modelRouting = modelRouting ?? {};
+    }
+    /**
+     * Phase 2.4 — substitute the caller-supplied model with a
+     * configured tier when `required_capabilities` asks for one.
+     * Returns the caller's model unchanged when no matching tier is
+     * configured, so the call is a no-op for users who haven't set
+     * up routing.
+     */
+    applyCapabilityRouting(model, capabilities) {
+        if (!capabilities || capabilities.length === 0)
+            return model;
+        if (capabilities.includes('fast') && this.modelRouting.fast) {
+            return this.modelRouting.fast;
+        }
+        if (capabilities.includes('heavy') && this.modelRouting.heavy) {
+            return this.modelRouting.heavy;
+        }
+        if (this.modelRouting.default) {
+            return this.modelRouting.default;
+        }
+        return model;
     }
     resolveDirectConfig(model) {
         const modelLower = model.toLowerCase();
@@ -241,29 +297,39 @@ export class AgentClient {
         return null;
     }
     /**
-     * Maps a model id to the provider that will actually serve the
-     * request — used as the key for `providerCooldown` tracking. The
-     * `freellmapi` sentinel covers the proxy path; everything else
-     * routes through a direct provider.
+     * Maps a model id to the tracking key for `providerCooldown`.
+     * Model-specific isolation ensures a timeout on one model (e.g.
+     * `openrouter:claude-3`) does not poison other models on the
+     * same provider gateway.
      */
-    getProviderId(model) {
+    getCooldownKey(model) {
         const direct = this.resolveDirectConfig(model);
         if (direct)
-            return direct.providerName;
-        return 'freellmapi';
+            return `${direct.providerName}:${model}`;
+        return `freellmapi:${model}`;
     }
     /* ─── Non-streaming chat ─── */
     async chat(messages, model, options = {}) {
         const { signal: externalSignal, ...restOptions } = options;
-        const providerId = this.getProviderId(model);
-        providerCooldown.assertAvailable(providerId);
+        // Phase 2.4 — substitute the model BEFORE provider resolution so
+        // both the routing decision and the eventual request body see
+        // the same name. No-op when no capabilities are tagged or no
+        // tier is configured.
+        model = this.applyCapabilityRouting(model, options.required_capabilities);
+        const cooldownKey = this.getCooldownKey(model);
+        providerCooldown.assertAvailable(cooldownKey);
         const direct = this.resolveDirectConfig(model);
         const isAnthropicDirect = direct && direct.providerName === 'anthropic';
-        // Combine external abort signal with the internal 60s timeout so
-        // the request aborts on EITHER signal (timeout OR user cancellation).
-        const combinedSignal = externalSignal
-            ? AbortSignal.any([AbortSignal.timeout(60000), externalSignal])
-            : AbortSignal.timeout(60000);
+        // Direct-mode safety: refuse to silently fall through to the
+        // FreeLLMAPI proxy when the user explicitly chose direct mode at
+        // setup. A user who picked direct deserves a loud error, not a
+        // request that surprises them by transiting a third-party SaaS.
+        if (this.providerMode === 'direct' && !direct) {
+            throw new DirectModelUnresolvedError(model);
+        }
+        // The timeout was removed to allow slow reasoning models to take as long as they need.
+        // The request will only abort if the user explicitly cancels it via `externalSignal`.
+        const combinedSignal = externalSignal;
         let requestUrl = `${this.baseUrl}/chat/completions`;
         let headers = {
             'Content-Type': 'application/json',
@@ -335,7 +401,7 @@ export class AgentClient {
             body = JSON.stringify(bodyObj);
         }
         // Check for pre-flight cancellation
-        if (combinedSignal.aborted) {
+        if (combinedSignal?.aborted) {
             throw new Error('Task cancelled by user.');
         }
         let lastError = null;
@@ -356,7 +422,7 @@ export class AgentClient {
                 }
                 // Retryable errors
                 if (RETRYABLE_STATUS_CODES.has(response.status)) {
-                    trackProviderError(providerId, response.status, `HTTP ${response.status}`);
+                    trackProviderError(cooldownKey, response.status, `HTTP ${response.status}`);
                     const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                     if (attempt < MAX_RETRIES) {
                         console.log(`${colors.yellow}⚠  [API] Error ${response.status}. Retrying in ${(delayMs / 1000).toFixed(1)}s (${attempt + 1}/${MAX_RETRIES})${colors.reset}`);
@@ -369,9 +435,11 @@ export class AgentClient {
                     throw new Error(`API error (${response.status}): ${errorText}`);
                 }
                 const rawData = await response.json();
-                const data = isAnthropicDirect ? translateAnthropicToOpenAI(rawData) : rawData;
+                const data = isAnthropicDirect
+                    ? translateAnthropicToOpenAI(rawData)
+                    : rawData;
                 const choice = data.choices[0];
-                providerCooldown.recordSuccess(providerId);
+                providerCooldown.recordSuccess(cooldownKey);
                 // ChatResult.content is `string | null`. The widened
                 // ChatMessage.content union allows blocks on input, but
                 // every provider we ship returns text-only assistant
@@ -427,7 +495,7 @@ export class AgentClient {
                     }
                 }
                 if (isNetworkError && attempt < MAX_RETRIES) {
-                    trackProviderError(providerId, 0, lastError.message.slice(0, 200));
+                    trackProviderError(cooldownKey, 0, lastError.message.slice(0, 200));
                     const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                     console.log(`${colors.yellow}⚠  [Network] ${lastError.message.slice(0, 60)}. Retrying in ${(delayMs / 1000).toFixed(1)}s (${attempt + 1}/${MAX_RETRIES})${colors.reset}`);
                     await sleep(delayMs);
@@ -442,9 +510,9 @@ export class AgentClient {
         throw lastError ?? new Error('All retry attempts exhausted.');
     }
     /* ─── Streaming chat (SSE) ─── */
-    async *executeSingleChatStreamAttempt(requestUrl, headers, body, model, isAnthropicDirect, signal = AbortSignal.timeout(60000)) {
+    async *executeSingleChatStreamAttempt(requestUrl, headers, body, model, isAnthropicDirect, signal) {
         // Pre-flight cancellation check
-        if (signal.aborted) {
+        if (signal?.aborted) {
             throw new Error('Task cancelled by user.');
         }
         const response = await fetch(requestUrl, {
@@ -722,13 +790,18 @@ export class AgentClient {
     }
     async *chatStream(messages, model, options = {}) {
         const { signal: externalSignal, ...restOptions } = options;
-        const combinedSignal = externalSignal
-            ? AbortSignal.any([AbortSignal.timeout(60000), externalSignal])
-            : AbortSignal.timeout(60000);
-        const providerId = this.getProviderId(model);
-        providerCooldown.assertAvailable(providerId);
+        // Phase 2.4 — capability-tier substitution (see chat() comment).
+        model = this.applyCapabilityRouting(model, options.required_capabilities);
+        // The timeout was removed to allow slow reasoning models to take as long as they need.
+        const combinedSignal = externalSignal;
+        const cooldownKey = this.getCooldownKey(model);
+        providerCooldown.assertAvailable(cooldownKey);
         const direct = this.resolveDirectConfig(model);
         const isAnthropicDirect = !!(direct && direct.providerName === 'anthropic');
+        // Same direct-mode safety as `chat()` — refuse to leak to proxy.
+        if (this.providerMode === 'direct' && !direct) {
+            throw new DirectModelUnresolvedError(model);
+        }
         let requestUrl = `${this.baseUrl}/chat/completions`;
         let headers = {
             'Content-Type': 'application/json',
@@ -807,7 +880,7 @@ export class AgentClient {
                     hasYielded = true;
                     yield chunk;
                 }
-                providerCooldown.recordSuccess(providerId);
+                providerCooldown.recordSuccess(cooldownKey);
                 return; // Success — don't retry
             }
             catch (error) {
@@ -850,7 +923,7 @@ export class AgentClient {
                     }
                 }
                 if (lastError instanceof HttpError && RETRYABLE_STATUS_CODES.has(lastError.status)) {
-                    trackProviderError(providerId, lastError.status, `HTTP ${lastError.status}`);
+                    trackProviderError(cooldownKey, lastError.status, `HTTP ${lastError.status}`);
                     const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                     if (attempt < MAX_RETRIES) {
                         console.log(`${colors.yellow}⚠  [API] Error ${lastError.status}. Retrying in ${(delayMs / 1000).toFixed(1)}s (${attempt + 1}/${MAX_RETRIES})${colors.reset}`);
@@ -859,7 +932,7 @@ export class AgentClient {
                     }
                 }
                 if (isNetworkError && attempt < MAX_RETRIES) {
-                    trackProviderError(providerId, 0, lastError.message.slice(0, 200));
+                    trackProviderError(cooldownKey, 0, lastError.message.slice(0, 200));
                     const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                     console.log(`${colors.yellow}⚠  [Network] ${lastError.message.slice(0, 60)}. Retrying in ${(delayMs / 1000).toFixed(1)}s (${attempt + 1}/${MAX_RETRIES})${colors.reset}`);
                     await sleep(delayMs);
@@ -979,8 +1052,8 @@ export class AgentClient {
         }
     }
     async getEmbedding(text, model = 'text-embedding-3-small') {
-        const providerId = this.getProviderId(model);
-        providerCooldown.assertAvailable(providerId);
+        const cooldownKey = this.getCooldownKey(model);
+        providerCooldown.assertAvailable(cooldownKey);
         const direct = this.resolveDirectConfig(model);
         let requestUrl = `${this.baseUrl}/embeddings`;
         let headers = {
@@ -1008,7 +1081,7 @@ export class AgentClient {
                     body,
                 });
                 if (RETRYABLE_STATUS_CODES.has(response.status)) {
-                    trackProviderError(providerId, response.status, `HTTP ${response.status}`);
+                    trackProviderError(cooldownKey, response.status, `HTTP ${response.status}`);
                     const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                     if (attempt < MAX_RETRIES) {
                         if (this.verbose) {
@@ -1024,7 +1097,7 @@ export class AgentClient {
                 }
                 const data = await response.json();
                 if (data.data && data.data[0] && data.data[0].embedding) {
-                    providerCooldown.recordSuccess(providerId);
+                    providerCooldown.recordSuccess(cooldownKey);
                     return data.data[0].embedding;
                 }
                 throw new Error('Malformed embedding response structure');
@@ -1037,7 +1110,7 @@ export class AgentClient {
                     error.message.includes('fetch failed') ||
                     error.message.includes('ETIMEDOUT'));
                 if (isNetworkError) {
-                    trackProviderError(providerId, 0, error.message.slice(0, 200));
+                    trackProviderError(cooldownKey, 0, error.message.slice(0, 200));
                 }
                 const delayMs = BASE_DELAY_MS * Math.pow(2, attempt);
                 await sleep(delayMs);
@@ -1153,6 +1226,11 @@ function messagesForOpenAIWire(messages) {
 }
 function translateOpenAIToAnthropic(messages, model, options) {
     let system = '';
+    // Phase 4.6 — `any[]` paydown. The shape here is the Anthropic
+    // wire-format messages array. The narrower type doesn't capture
+    // every field the SDK accepts (tool_result, document blocks),
+    // but `unknown[]` lets us keep type-safety at this construction
+    // site without inventing a half-typed interface that drifts.
     const anthropicMessages = [];
     for (const msg of messages) {
         if (msg.role === 'system') {
@@ -1173,6 +1251,7 @@ function translateOpenAIToAnthropic(messages, model, options) {
         else if (msg.role === 'assistant') {
             const assistantText = extractTextFromContent(msg.content);
             if (msg.tool_calls && msg.tool_calls.length > 0) {
+                // Phase 4.6 — see `anthropicMessages` comment for rationale.
                 const contentBlocks = [];
                 if (assistantText.length > 0) {
                     contentBlocks.push({ type: 'text', text: assistantText });
@@ -1279,7 +1358,7 @@ function translateAnthropicToOpenAI(anthropicRes) {
             role: 'assistant',
             content: text || null,
         },
-        finish_reason: finishReasonMap[anthropicRes.stop_reason] || 'stop',
+        finish_reason: finishReasonMap[anthropicRes.stop_reason ?? ''] || 'stop',
     };
     if (toolCalls.length > 0) {
         choice.message.tool_calls = toolCalls;