manbot/_board/TASKS.md

Enhanced HTTP Get Tool Tasks

Phase 1: Dependencies and Configuration

Task 1.1: Add Required Dependencies

File: package.json Dependencies: None Description: Add Playwright, stealth plugin, and HTML-to-Markdown conversion libraries. Acceptance Criteria:

• Add playwright to dependencies
• Add playwright-extra to dependencies
• Add puppeteer-extra-plugin-stealth to dependencies
• Add turndown to dependencies
• Add @types/turndown to devDependencies
• Run npm install successfully
• All dependencies are compatible with Node.js 20+

Task 1.2: Install Playwright Browsers

Dependencies: Task 1.1 Description: Install Chromium browser for Playwright. Acceptance Criteria:

• Run npx playwright install chromium
• Verify Chromium is installed successfully
• Document browser installation in README or setup docs

Task 1.3: Add Browser Service Configuration

File: config.json Dependencies: None Description: Add configuration section for browser service. Acceptance Criteria:

• Add browserService object with headless, timeout, enableStealth, reuseContext properties
• Set sensible defaults (headless: true, timeout: 30000, enableStealth: true, reuseContext: true)
• Configuration is valid JSON

Task 1.4: Update Config TypeScript Types

File: src/shared/config.ts Dependencies: Task 1.3 Description: Add TypeScript types and validation for browser service configuration. Acceptance Criteria:

• Add BrowserServiceConfig interface with typed properties
• Add browserService to main config type
• Add validation for browser service config in getConfig()
• TypeScript compilation succeeds with no errors

---

Phase 2: Core Utilities

Task 2.1: Create HTML to Markdown Converter

File: src/utils/html-to-markdown.ts Dependencies: Task 1.1 Description: Create utility to convert HTML to clean Markdown using Turndown. Acceptance Criteria:

• Exports htmlToMarkdown(html: string, options?: ConversionOptions): string
• Configures Turndown to preserve links, images, code blocks, tables, headings, lists
• Strips scripts, styles, navigation, footers, and other non-content elements
• Handles malformed HTML gracefully
• Returns empty string for empty/invalid input
• Includes JSDoc documentation

Task 2.2: Add HTML to Markdown Tests

File: src/utils/__tests__/html-to-markdown.test.ts Dependencies: Task 2.1 Description: Create unit tests for HTML to Markdown converter. Acceptance Criteria:

• Tests conversion of headings (h1-h6)
• Tests conversion of lists (ul, ol)
• Tests conversion of links and images
• Tests conversion of code blocks and inline code
• Tests conversion of tables
• Tests stripping of scripts and styles
• Tests handling of malformed HTML
• Tests handling of empty input
• All tests pass with npm test

Task 2.3: Create Browser Configuration

File: src/services/browser-config.ts Dependencies: None Description: Create configuration for user agents, viewports, and stealth settings. Acceptance Criteria:

• Exports array of 10+ realistic user agents (Chrome, Firefox, Safari on Windows, macOS, Linux)
• Exports array of common viewport sizes (1920x1080, 1366x768, 1536x864, etc.)
• Exports function getRandomUserAgent(): string
• Exports function getRandomViewport(): { width: number, height: number }
• Exports stealth plugin configuration object
• Includes JSDoc documentation explaining bot detection bypass techniques

---

Phase 3: Browser Service

Task 3.1: Create Browser Service Core

File: src/services/browser-service.ts Dependencies: Task 1.1, Task 1.4, Task 2.3 Description: Create service to manage Playwright browser instances. Acceptance Criteria:

• Implements singleton pattern for browser instance
• Exports BrowserService class extending BaseProcess
• Implements fetchWithBrowser(url: string, options?: BrowserFetchOptions): Promise<BrowserFetchResult>
• Implements close(): Promise<void> for cleanup
• Configures Chromium with stealth plugin
• Uses random user agent and viewport for each request
• Handles browser launch errors gracefully
• Includes timeout handling (from config)
• Reuses browser context when reuseContext is enabled

Task 3.2: Add Realistic Behavior to Browser Service

File: src/services/browser-service.ts Dependencies: Task 3.1 Description: Add human-like behaviors to bypass bot detection. Acceptance Criteria:

• Adds random delay (100-500ms) before page interaction
• Implements realistic mouse movement to random coordinates
• Waits for network idle before extracting content
• Adds random scroll behavior for long pages
• Configures browser to disable automation flags
• Sets realistic browser headers (Accept-Language, Accept-Encoding, etc.)

Task 3.3: Add Browser Service Tests

File: src/services/__tests__/browser-service.test.ts Dependencies: Task 3.2 Description: Create integration tests for browser service. Acceptance Criteria:

• Tests fetching a simple static HTML page
• Tests fetching a page with JavaScript (mocked SPA)
• Tests timeout handling with slow-loading page
• Tests browser instance reuse
• Tests cleanup on shutdown
• Tests error handling for invalid URLs
• Tests stealth plugin is applied
• All tests pass with npm test

---

Phase 4: Enhanced HTTP Get Tool

Task 4.1: Update HTTP Get Tool with Smart Fallback

File: src/services/tool-host.ts Dependencies: Task 2.1, Task 3.2 Description: Enhance httpGetTool to support Playwright with smart fallback logic. Acceptance Criteria:

• Accepts new optional parameters: useBrowser?: boolean, convertToMarkdown?: boolean
• Implements fallback logic: try fetch first, use Playwright on 403/401 or if useBrowser is true
• Detects HTML content type from response headers
• Converts HTML to Markdown when convertToMarkdown is true (default for HTML)
• Returns enhanced response with status, body, contentType, finalUrl, method fields
• Handles errors from both fetch and Playwright
• Logs which method was used (fetch vs browser)

Task 4.2: Add HTTP Get Tool Tests

File: src/services/__tests__/tool-host.test.ts Dependencies: Task 4.1 Description: Create integration tests for enhanced HTTP get tool. Acceptance Criteria:

• Tests successful fetch with simple URL
• Tests fallback to Playwright on 403 response
• Tests explicit useBrowser: true parameter
• Tests HTML to Markdown conversion
• Tests response format includes all required fields
• Tests error handling for invalid URLs
• Tests error handling for network failures
• All tests pass with npm test

---

Phase 5: Planner Integration

Task 5.1: Update Planner Prompt

File: src/agents/prompts/planner.ts Dependencies: Task 4.1 Description: Update planner prompt to document enhanced http_get capabilities. Acceptance Criteria:

• Updates http_get tool documentation to include useBrowser and convertToMarkdown parameters
• Adds example showing when to use useBrowser: true (e.g., for SPAs, sites with bot detection)
• Adds example showing Markdown conversion for web scraping
• Documents that HTML responses are automatically converted to Markdown
• Explains that fetch is tried first for performance, with automatic fallback to browser

---

Phase 6: Testing & Documentation

Task 6.1: Manual End-to-End Testing

Dependencies: All previous tasks Description: Perform manual testing of the complete enhanced HTTP get flow. Acceptance Criteria:

• Test fetching static HTML page (should use fetch)
• Test fetching SPA website (should fallback to Playwright)
• Test fetching site with bot detection (should use Playwright with stealth)
• Test explicit useBrowser: true parameter
• Test Markdown conversion quality on real websites
• Test error handling with invalid URLs
• Test timeout handling with slow sites
• Verify logs show which method was used
• Verify response times (fetch should be faster than Playwright)

Task 6.2: Performance Benchmarking

Dependencies: Task 6.1 Description: Benchmark performance of fetch vs Playwright. Acceptance Criteria:

• Measure average response time for fetch (should be <1s for most sites)
• Measure average response time for Playwright (should be <5s for most sites)
• Measure browser startup time (first request vs subsequent requests)
• Document performance characteristics in code comments or README
• Verify browser context reuse improves performance

Task 6.3: Update README Documentation

File: README.md Dependencies: Task 6.1 Description: Document the enhanced HTTP get tool in the README. Acceptance Criteria:

• Adds section explaining enhanced http_get capabilities
• Documents when Playwright is used vs fetch
• Documents bot detection bypass techniques
• Documents HTML to Markdown conversion
• Includes examples of using useBrowser parameter
• Documents browser installation requirement (npx playwright install chromium)

Task 6.4: Add Troubleshooting Guide

File: README.md or docs/TROUBLESHOOTING.md Dependencies: Task 6.3 Description: Create troubleshooting guide for common browser service issues. Acceptance Criteria:

• Documents how to debug Playwright issues (headless: false, slowMo, screenshots)
• Documents common bot detection bypass failures and solutions
• Documents how to handle sites with CAPTCHA
• Documents browser installation issues
• Documents timeout configuration
• Documents how to view browser logs

---

Replace read_file and write_file with shell Tool Tasks

Phase 1: Core Shell Tool Implementation

Task S1.1: Implement Shell Tool Core

File: src/services/tool-host.ts Dependencies: None Description: Implement the core shell tool that executes shell commands. Acceptance Criteria:

• Add shellTool method that accepts command (required, string) and cwd (optional, string)
• Use Node.js child_process.exec or spawn to execute commands
• Default cwd to sandboxDir from config
• Return structured response: { stdout, stderr, exitCode, command, cwd }
• Handle command execution errors gracefully
• Include JSDoc documentation

Task S1.2: Add Command Validation

File: src/services/tool-host.ts Dependencies: Task S1.1 Description: Add validation to ensure commands operate within sandbox. Acceptance Criteria:

• Add validateCommand(command: string, cwd: string): { allowed: boolean, reason?: string } method
• Validate that cwd is within sandbox directory (resolve and check)
• Validate that resolved cwd path starts with sandboxDir and doesn't contain ..
• Return clear error messages when validation fails
• Handle edge cases (empty cwd, relative paths, etc.)

Task S1.3: Replace read_file and write_file Registration

File: src/services/tool-host.ts Dependencies: Task S1.2 Description: Update tool registration to use shell tool instead of read_file/write_file. Acceptance Criteria:

• Remove readFileTool and writeFileTool from registerDefaultTools()
• Add shell tool registration: this.tools.set("shell", this.shellTool.bind(this))
• Remove readFileTool and writeFileTool method implementations
• Remove resolvePath method if no longer needed (or adapt for shell tool validation)
• Update file header comment to reflect shell tool instead of read_file/write_file

Task S1.4: Add Shell Tool Tests

File: src/services/__tests__/tool-host.test.ts Dependencies: Task S1.3 Description: Create tests for shell tool functionality. Acceptance Criteria:

• Test reading file: cat file.txt command
• Test writing file: echo "content" > file.txt then verify file exists
• Test listing files: ls -la command
• Test custom cwd parameter
• Test sandbox path validation (reject paths outside sandbox)
• Test invalid command handling
• Test command with non-zero exit code
• Test stdout and stderr capture
• All tests pass with npm test

---

Phase 2: Update Dependent Services

Task S2.1: Update Generator Service Content Extraction

File: src/services/generator-service.ts Dependencies: Task S1.3 Description: Update generator service to extract content from shell tool responses. Acceptance Criteria:

• Remove read_file response handling (content extraction logic)
• Add shell tool response handling:
  • Extract stdout from shell tool responses
  • Include stderr in context if non-empty (for debugging)
  • Handle read operations (e.g., cat file.txt outputs to stdout)
• Update comments to reflect shell tool usage
• Test that shell tool stdout is correctly extracted and used in prompts

Task S2.2: Add Generator Service Tests

File: src/services/__tests__/generator-service.test.ts Dependencies: Task S2.1 Description: Test generator service handles shell tool responses correctly. Acceptance Criteria:

• Test that shell tool stdout is extracted correctly
• Test that stderr is handled appropriately (included if non-empty)
• Test that read operations (cat) produce expected content in prompts
• All tests pass with npm test

---

Phase 3: Update Planner and Documentation

Task S3.1: Update Planner Prompt - Remove Old Tools

File: src/agents/prompts/planner.ts Dependencies: Task S1.3 Description: Remove read_file and write_file from planner prompt tool list. Acceptance Criteria:

• Remove read_file and write_file from available tools list
• Update tool list to show: shell, http_get, http_search
• Update "DO NOT invent tool names" warning to reflect new tool list
• Remove examples that use read_file/write_file

Task S3.2: Update Planner Prompt - Document Shell Tool

File: src/agents/prompts/planner.ts Dependencies: Task S3.1 Description: Add comprehensive documentation for shell tool. Acceptance Criteria:

• Add shell tool documentation section with:
  • Purpose: Execute shell commands for file operations, process management, etc.
  • Arguments: command (required), cwd (optional)
  • Response format: { stdout, stderr, exitCode, command, cwd }
  • Common use cases:
    • Read file: cat path/to/file.txt
    • Write file: echo "content" > path/to/file.txt or heredoc syntax
    • List files: ls -la, ls -la directory/
    • Check processes: ps aux, pgrep process_name
    • Search files: grep "pattern" file.txt, find . -name "*.txt"
  • Security note: All file operations restricted to sandbox directory
• Add examples showing shell tool usage for file operations
• Update "For file operations" section to use shell tool

Task S3.3: Update Config Documentation

File: src/shared/config.ts Dependencies: Task S1.3 Description: Update config comments to reflect shell tool usage. Acceptance Criteria:

• Update ToolHostConfig interface comment:
  • Change from: /** Directory allowed for read_file/write_file. Paths outside are rejected. */
  • Change to: /** Directory allowed for shell tool file operations. Paths outside are rejected. */
• TypeScript compilation succeeds with no errors

---

Phase 4: Update External Documentation

Task S4.1: Update README.md

File: README.md Dependencies: Task S3.2 Description: Update README to reflect shell tool instead of read_file/write_file. Acceptance Criteria:

• Update services list: Replace read_file, write_file with shell tool
• Document shell tool capabilities and common use cases
• Add example shell commands for file operations
• Update any references to old tools

Task S4.2: Update AI-Agent.md

File: AI-Agent.md Dependencies: Task S3.2 Description: Update AI-Agent.md to reflect shell tool. Acceptance Criteria:

• Update tool-host.ts description: Replace read_file, write_file with shell
• Update tool capabilities list
• Update any architecture diagrams or descriptions

---

Phase 5: Testing and Verification

Task S5.1: Manual Testing - File Operations

Dependencies: Task S1.4, Task S2.2 Description: Manually test file read/write operations using shell tool. Acceptance Criteria:

• Test reading file: Execute cat test.txt and verify stdout contains file contents
• Test writing file: Execute echo "test content" > test.txt and verify file created
• Test appending: Execute echo "more content" >> test.txt and verify appended
• Test listing: Execute ls -la and verify directory listing in stdout
• Test nested directories: Execute ls -la subdir/ and verify works correctly
• Verify all operations respect sandbox directory

Task S5.2: Manual Testing - Sandbox Enforcement

Dependencies: Task S1.4 Description: Test that sandbox restrictions are properly enforced. Acceptance Criteria:

• Try to access file outside sandbox: cat /etc/passwd (if sandbox is not root)
• Verify command is rejected or fails appropriately
• Try to change directory outside sandbox: cd / && pwd
• Verify sandbox restrictions are enforced
• Try relative path traversal: cat ../../etc/passwd
• Verify path traversal is blocked
• Test that operations within sandbox work correctly

Task S5.3: Manual Testing - Process Management

Dependencies: Task S1.4 Description: Test shell tool with process management commands. Acceptance Criteria:

• Execute ps aux | head -5 and verify output shows process list
• Execute pgrep node and verify output shows Node.js process IDs
• Execute echo $PATH and verify environment variables are accessible
• Test command chaining: ls -la | grep ".txt"
• Verify stdout/stderr are captured correctly

Task S5.4: Manual Testing - Error Handling

Dependencies: Task S1.4 Description: Test error handling for various failure scenarios. Acceptance Criteria:

• Execute invalid command: nonexistentcommand
• Verify stderr contains error message
• Verify exit code is non-zero
• Verify structured error response
• Test command that fails: cat nonexistentfile.txt
• Verify stderr contains file not found error
• Verify exit code reflects failure
• Test timeout handling (if implemented)

Task S5.5: End-to-End Integration Test

Dependencies: All previous tasks Description: Test complete flow from planner to executor using shell tool. Acceptance Criteria:

• Send task via Telegram: "Read the file test.txt and summarize it"
• Verify planner generates plan using shell tool (cat test.txt)
• Verify executor executes shell tool correctly
• Verify generator service extracts stdout and uses it in prompt
• Verify final response includes file content summary
• Test write operation: "Write 'Hello World' to hello.txt"
• Verify file is created correctly
• Verify response confirms file creation

---

Ollama Model Manager Tasks

Phase M1: Core Infrastructure

Task M1.1: Enhance OllamaAdapter with Warmup Support

File: src/services/ollama-adapter.ts Dependencies: None Description: Add a warmup method to OllamaAdapter that uses the /api/chat endpoint with a minimal prompt and supports the keep_alive parameter. Acceptance Criteria:

• warmup(model: string, keepAlive: string | number): Promise<void> implemented.
• Uses /api/chat with stream: false.
• chat and generate methods updated to accept keep_alive in options.
• Error handling for warmup failures.

Task M1.2: Add Model Manager Configuration

Files: config.json, src/shared/config.ts Dependencies: None Description: Add configuration settings for the model manager, including keep-alive durations and warmup prompts. Acceptance Criteria:

• modelManager section added to config.json.
• TypeScript types and validation added to src/shared/config.ts.
• Settings include largeModelKeepAlive and warmupPrompt.

Task M1.3: Implement ModelManagerService Core

File: src/services/model-manager.ts Dependencies: Task M1.1, Task M1.2 Description: Create the ModelManagerService to manage tiered model lifecycles and ensure model availability. Acceptance Criteria:

• ModelManagerService class created.
• ensureModelLoaded(tier: ModelTier) implemented with concurrency safety (locking).
• prewarmModels() implemented for sequential loading of small and medium models.
• Correct keep_alive values passed during warmup based on tier.

Task M1.4: Unit Tests for ModelManagerService

File: src/services/__tests__/model-manager.test.ts Dependencies: Task M1.3 Description: Create comprehensive unit tests for the ModelManagerService. Acceptance Criteria:

• Tests sequential prewarming logic.
• Tests concurrency safety (concurrent ensureModelLoaded calls).
• Tests correct tier-to-model mapping and keep-alive parameters.
• All tests pass with npm test.

---

Phase M2: Integration & Verification

Task M2.1: Integrate ModelManager into GeneratorService

File: src/services/generator-service.ts Dependencies: Task M1.3 Description: Update GeneratorService to call the ModelManagerService before performing any inference. Acceptance Criteria:

• ModelManagerService injected into GeneratorService.
• ensureModelLoaded called before ollama.chat and ollama.generate.
• Works correctly with all model tiers.

---

Phase P8: Natural Language Analysis of Tool Outputs

Overview

Improve LLM outputs by ensuring that data gathered from tools (search, shell, web) is analyzed and presented as coherent text. This phase introduces a "Narrative Rule" for the planner and a dedicated analyzer service role.

Proposed Changes

Component 1: Analyzer Prompt

• Create a specialized analyzer system prompt that forbids raw JSON in final outputs.
• Implement a user prompt template that clearly separates "User Goal" from "Raw Tool Data".

Component 2: Planner "Narrative Rule"

• Update the Planner Agent to always conclude research tasks with a generate_text node.
• These nodes are tagged with system_prompt: "analyzer".

Component 3: Targeted Deployment in Generator

• The Generator Service detects the "analyzer" tag and applies the synthesis prompt.
• This prevents breaking internal tool chains that require JSON.

Task M2.2: Implement Startup Prewarming in Orchestrator

File: src/core/orchestrator.ts Dependencies: Task M1.3 Description: Trigger the prewarming of small and medium models during application startup. Acceptance Criteria:

• ModelManagerService initialized in Orchestrator.
• prewarmModels() called during the bootstrap phase.
• Prewarming does not block the main application flow.

Task M2.3: Integration Testing for Inference Flow

File: src/services/__tests__/generator-model-manager.test.ts Dependencies: Task M2.1 Description: Integration test to verify that inference requests correctly trigger model loading. Acceptance Criteria:

• Mock Ollama adapter.
• Verify GeneratorService + ModelManagerService interaction.
• All tests pass with npm test.

Task M2.4: Manual Verification and Documentation

Dependencies: All previous tasks Description: Perform manual verification of model states and update project documentation. Acceptance Criteria:

• Verify models are loaded at startup using ollama ps.
• Verify large model loads on demand and unloads after inactivity.
• Update README.md or AI-Agent.md with model management details.

---

Phase 8: Natural Language Analysis

Task 8.1: Implement Analyzer Prompt and Synthesis Logic

Files: src/agents/prompts/analyzer.ts, src/services/generator-service.ts Description: Create the analyzer role and integrate it into the generator service. Acceptance Criteria:

• ANALYZER_SYSTEM_PROMPT created.
• GeneratorService swaps "analyzer" tag for full prompt.
• GeneratorService uses analysis-specific user template.

Task 8.2: Planner Narrative Rule and Search Examples

File: src/agents/prompts/planner.ts Description: Update the planner to ensure it always schedules an analysis node after data-gathering tools. Acceptance Criteria:

• Narrative Rule added to system prompt.
• Search & Answer few-shot example added.
• All research plans end with a generate_text node tagged as "analyzer".

Phase P9: Dynamic Skills System Tasks

Phase 9.1: Core Infrastructure

Task 9.1: Add Skills Configuration

File: src/shared/config.ts Dependencies: None Description: Add skills section to the central configuration system to specify the skills directory. Acceptance Criteria:

• SkillsConfig interface added.
• skills added to AppConfig and DEFAULT_CONFIG.
• mergeEnv handles SKILLS_DIR environment variable.
• Configuration is successfully merged and validated.

Task 9.2: Implement SkillManager Service

File: src/services/skill-manager.ts Dependencies: Task 9.1 Description: Create a service to load and parse skill manifests and prompts from disk. Acceptance Criteria:

• SkillManager class created.
• listSkills() correctly parses CONFIG.md (tables and lists).
• getSkillPrompt(name) loads SKILL.md for a specific skill.
• Paths are resolved correctly relative to the skills directory.
• Error handling for missing files or parse errors.

Phase 9.2: Agent Integration

Task 9.3: Update Planner Prompts for Skills

File: src/agents/prompts/planner.ts Dependencies: Task 9.2 Description: Modify the planner system prompt to support dynamic skills and the skill node type. Acceptance Criteria:

• PlannerPromptOptions updated to include optional skills array.
• buildPlannerPrompt injects "AVAILABLE SKILLS" section if skills are provided.
• System prompt includes instructions for creating type: "skill" nodes.
• Example skill node provided in the prompt.

Task 9.4: Integrate SkillManager into PlannerAgent

File: src/agents/planner-agent.ts Dependencies: Task 9.3 Description: Wire the SkillManager into the PlannerAgent so it loads skills on every request. Acceptance Criteria:

• SkillManager initialized in PlannerAgent constructor.
• handleEnvelope calls listSkills() and passes them to buildPlannerPrompt.
• Skills are successfully injected into the LLM prompt.

Task 9.5: Support Skill Nodes in ExecutorAgent

File: src/agents/executor-agent.ts Dependencies: Task 9.2 Description: Update the executor to recognize and process the new skill node type. Acceptance Criteria:

• SkillManager initialized in ExecutorAgent.
• TYPE_TO_SERVICE maps skill to model-router.
• dispatchNode handles type: skill by loading the prompt via SkillManager.
• Skill task and prompt are correctly mapped to model-router input.

Phase 9.3: Validation and Demo

Task 9.6: Create Demo Skill

File: skills/demo/SKILL.md, skills/CONFIG.md Dependencies: Task 9.4, Task 9.5 Description: Create a sample skill to verify the system end-to-end. Acceptance Criteria:

• skills/CONFIG.md initialized with a demo skill entry.
• skills/demo/SKILL.md created with clear instructions.
• Verification that the skill appears in the planner's context.

Task 9.7: Project Compilation

File: Project-wide Dependencies: All Phase 9 tasks Description: Ensure the project builds successfully with the new changes. Acceptance Criteria:

• npm run build succeeds without TypeScript errors.
• Lint errors related to the new services are resolved.

Task 9.8: Prioritize Skill Usage in Planner

File: src/agents/prompts/planner.ts Dependencies: Task 9.3 Description: Update the planner prompts to ensure that skills are given higher priority than raw tool usage. Acceptance Criteria:

• Added ## 💎 SKILLS PRECEDENCE (HIGHEST PRIORITY) section to PLANNER_SYSTEM_PROMPT.
• Updated buildPlannerPrompt to label the skills section as HIGHEST PRIORITY and added a STRICT RULE instruction.
• Verified that instructions tell the model to check skills BEFORE raw tools.

Phase P10: Tool-Aware Active Skills Tasks

Phase 10.1: Alignment and Planning

Task 10.1: Planner "Dependency Rule" for Skills

File: src/agents/prompts/planner.ts Dependencies: Task 9.8 Description: Instruct the planner to provide tool outputs to skills that need them. Acceptance Criteria:

• Added instructions to PLANNER_SYSTEM_PROMPT about skill dependencies.
• Updated buildPlannerPrompt to emphasize that skills don't have direct tool access; they need "data nodes" as dependencies.

Task 10.2: Refine Skill Prompts for Tool Dependencies

File: skills/time/SKILL.md Dependencies: Task 10.1 Description: Update the "time" skill to expect input from a shell tool rather than promising to run it itself. Acceptance Criteria:

• Updated time skill prompt to explain it formats raw date output.

Phase 10.2: Active Skill Execution (Future)

Task 10.3: Active Skill Research and Implementation

File: src/agents/executor-agent.ts, src/services/generator-service.ts Dependencies: Task 9.5 Description: Investigate and implement a way for skill nodes to perform their own tool calls if the model supports it. Acceptance Criteria:

• Defined protocol for tool-calling within the GeneratorService.

Research Skill Tasks

Phase 11: Core Research Skill Implementation

Task SK-RS-01: Verify Lynx Dependency in Orchestrator

File: src/core/orchestrator.ts Description: Implement a check at startup to ensure lynx is installed on the system. Acceptance Criteria:

• Add verifySystemDependencies() method.
• Check lynx --version or command -v lynx.
• Log an error or warning if missing.
• Orchestrator continues to run but skill functionality is flagged as degraded.

Task SK-RS-02: Create Research Skill Prompt

File: skills/research/SKILL.md Description: Create the instruction file for the research skill. Acceptance Criteria:

• Define DuckDuckGo search URL: https://html.duckduckgo.com/html?q={QUERY}.
• Instruct LLM to use lynx -dump for both searching and browsing.
• Provide clear instructions on link extraction from References section.
• Set recursive depth limits and summarization quality standards.

Task SK-RS-03: Register Research Skill in Manifest

File: skills/CONFIG.md Description: Register the new skill in the global skills configuration. Acceptance Criteria:

• Add research entry with description.
• Ensure the description highlights its capability for deep web searches.

Task SK-RS-04: Update Planner with Research Guidance

File: src/agents/prompts/planner.ts Description: Add a few-shot example to the planner to encourage the use of the research skill for information gathering. Acceptance Criteria:

• New example showing task -> search -> browse -> summarize.
• Instructions to prefer research over generic http_search for complex queries.

Task SK-RS-05: E2E Verification of Research Skill

Description: Perform full end-to-end testing of the research capability. Acceptance Criteria:

• Ask the agent complex questions requiring multi-step web browsing.
• Verify the agent follows links and consolidates findings.
• Verify the final response is informative and well-reasoned.