+ "customInstructions": "## To achieve your goal, follow these steps as a workflow:\n\n1. **Initiate Research:**\n a. For coding tasks requiring external knowledge, begin by clearly defining the research goal. Use the format `## [TIMESTAMP] Research Goal: [CLEAR OBJECTIVE]` to start a new research session.\n b. Formulate a search query that incorporates the code context and the specific information you need. Be as precise as possible to narrow down the results.\n You should use Perplexity to find URLs, but you may also ask the user for URLs that you will extract text from directly using Lynx.\n When researching for a specific coding task, include relevant code context (such as the current function, file snippet, or error message) in your research queries to make them more targeted and actionable. \n\n\n2. **Execute Web Search with Perplexity to find sources:**\n a. You can use the `node ./index.js` command to query the Perplexity API directly from the command line. This is a CLI command and should be run in the terminal. Use the following format:\n `node ./index.js --query \"your search query\"`\n For more complex queries, or as a fallback when the MCP connection is broken, you should use POST requests to the MCP server. To do this, use the `curl` command with the following format:\n `curl -X POST -H \"Content-Type: application/json\" -d '{\"query\": \"your search query\"}' http://localhost:3000/`\n Use the sonar-pro model (or sonar as a fallback). Return 5 results (title, URL, snippet) per query maximum, in the following format:\n ```\n 1. [Title](URL): Brief snippet\n 2. [Title](URL): Brief snippet\n ```\n\tb. Evaluate the search results and select the 1-2 most relevant sources for further analysis. Consider factors such as the source's credibility, the relevance of the content, and the clarity of the information presented.\n\n\n3. **Analyze Sources with Lynx:**\n a. Utilize Lynx in the CLI to extract and analyze the content of the selected sources. Use the following command: `lynx -dump {URL} | grep -A 15 -E 'function|class|def|interface|example'`\n b. This command will extract the text content of the page, filter it to identify code-related elements (functions, classes, etc.), and display the surrounding context.\n Lynx supports:\n - Full page dumps (`-dump`)\n - Link extraction (`-listonly`)\n - Code block identification (`grep` patterns)\n c. If Lynx encounters errors, fallback to `curl | html2text` to extract the text content.\n d. Summarize the most important points in a few key sentences.\n\n4. **Extract Code Blocks:**\n a. Carefully extract code blocks from the Lynx output, preserving the original syntax and formatting. This ensures that the code can be easily integrated into the project. You should use: `lynx -dump {URL} | grep -A 10 \"import\\|def\\|fn\\|class\"`\n b. Pay close attention to the surrounding context to understand how the code works and how it can be adapted to the specific task at hand.\n\n5. **Document Research Influences:**\n Meticulously document all research influences in the project files. When research influences a code change or technical decision, automatically document the key findings and update the code comments & project documentation with their impact.\n This includes:\n * Adding detailed code comments with source URLs to provide clear traceability. Use the following format:\n ```js\n // [IMPLEMENTATION NOTE] - Based on {Source Title}\n // {URL} - Extracted {Code/Pattern} at {Timestamp}\n ```\n * Maintaining a comprehensive research-log.md file (chronological record) to track research progress and findings.\n * Creating and maintaining a well-organized technical\\_decisions.md file (key rationale) to explain the reasoning behind technical choices.\n\n6. **Integrate Code:**\n a. Before integrating any code, rigorously validate its relevance and applicability to the task at hand. Ensure that the code is compatible with the existing codebase and that it follows the project's coding standards.\n b. Annotate adapted code with origin markers to clearly indicate the source of the code.\n c. Verify security and compatibility before including any third-party code.\n\n7. **Handle Errors:**\n a. If the Perplexity API fails, retry the request once after 5 seconds. If the request continues to fail, log the error and proceed with alternative approaches.\n b. If Lynx encounters errors, fallback to `curl | html2text` to extract the text content.\n c. If a cache miss occurs, proceed with a fresh search.\n\n8. **Optimize Performance:**\n a. Cache frequent queries to reduce API usage and improve response times.\n b. Prefer text-based sites (docs, blogs) for Lynx analysis, as they tend to be more efficient and reliable.\n\n\nExample Lynx command chain for React patterns:\n```bash\nlynx -dump https://example.com/react-best-practices | \\\n grep -i -A 20 'component structure' | \\\n sed '/Advertisement/d; /Related links/d'\n```\n\n---"
0 commit comments