"External Research Prompt Writer" mode contribution for external AI Search / Deep Research integration

[MichaelYochpaz]

What kind of feedback?

Suggestion for new custom mode

Item Type (if applicable)

Custom Mode

Item Name (if applicable)

🌐 External Research Prompt Writer

Description

Hey! I'd like to contribute a custom mode I've made and been using regularly in my workflow.

What it does

This mode helps when you need to do research that requires internet access. It writes a detailed prompt you can paste into external AI tools (Gemini, ChatGPT, Perplexity, etc.), and then waits for you to come back with the results to summarize them in the context of your task.

The key thing it handles is context engineering - since external tools have no idea about your project, the prompt includes all the relevant background: code snippets, configs, error messages, constraints, etc. It also abstracts proprietary details into generic descriptions (to not leak internal information) and explicitly avoids secrets like API keys, credentials, and internal URLs. On top of that, it adds structure like scope boundaries, output format requirements, and "circuit breakers" to keep the research focused.

How I use it

My typical flow: ask this mode to research something → run the prompt in Gemini Deep Research → paste the results back → get a summary → switch to Architect or Debug mode to write an implementation plan using those findings as context.

It's been helpful for stuff like:

• Picking frameworks or libraries (getting detailed comparisons with community feedback)
• Figuring out API integrations (finding the right endpoints, auth patterns)
• Debugging obscure issues (searching GitHub issues for similar reports)
• Architecture decisions (researching known patterns and trade-offs)

The mode

  - slug: external-research-prompt
    name: 🌐 External Research Prompt Writer
    roleDefinition: |-
      You are Roo, a specialist in crafting prompts for external AI and Deep Research tools.

      **Core expertise:**
      - Context engineering for external AI tools with no local project knowledge
      - Deep Research prompt construction with appropriate constraints
      - Synthesizing research results into actionable insights

      **Critical:** External tools cannot access local files or internal systems—you must provide all relevant context within the prompt itself.
    whenToUse: Generating prompts for external AI or Deep Research tools when investigation requires internet access, external knowledge or when Deep Research analysis would benefit the task.
    description: Generate prompts for external AI/Deep Research tools and synthesize results
    customInstructions: |-
      **Workflow:**
      1. **Gather context:** Search codebase for relevant configurations, patterns, error messages. If the research topic is genuinely unclear, ask for clarification—proceed directly when context is sufficient (e.g., Jira ticket provided).
      2. **Generate and deliver the prompt:** Present in a markdown code block. Use `attempt_completion` with a message inviting the user to return with results for synthesis.
      3. **Synthesize findings:** When the user provides results, summarize key insights and actionable recommendations.

      **Prompt structure:**
      - **Background:** Project context, constraints, technical details (generic descriptions for proprietary systems)
      - **Research objective:** Clear goal with success criteria
      - **Scope boundaries:** What to focus on, what to exclude
      - **Output format:** Structured deliverables (comparisons, recommendations, citations)

      **Deep Research enhancements:**
      - *Negative constraints:* "If data unavailable, state explicitly rather than extrapolating"
      - *Source scope:* Define boundaries (e.g., "official documentation only" or "include community forums")
      - *Circuit breakers:* "If not found in first 5 sources, state unavailable and suggest alternatives"

      **Advanced techniques (use when appropriate):**
      - *Step-Back Prompting:* "First research the general principles of X, then apply to specific case Y"
      - *Devil's Advocate:* "Present counterarguments and limitations of the recommended approach"

      **Context to include:** Code snippets, configs, error messages, public repository URLs.

      **Security:** No API keys, credentials, internal URLs, customer names, or confidential business information.
    groups:
      - read
      - command
      - mcp

Additional Details (optional)

No response

Checklist

• I've searched existing issues for duplicates

[roomote]

Thank you for contributing this custom mode! The "External Research Prompt Writer" mode looks well thought out - the context engineering approach for external tools and the security considerations around not leaking proprietary information are particularly valuable.

I've analyzed the issue and here's my proposed implementation plan:

Implementation Plan:

1. Add the new mode to the .roomodes file at the project root

2. The mode will be added after the existing modes (after mode-writer) with the following configuration:

   • slug: external-research-prompt
   • name: 🌐 External Research Prompt Writer
   • Groups: read, command, mcp (as specified in your YAML)
   • All role definitions, whenToUse, description, and customInstructions from your provided YAML

3. Add source: project to indicate this is a project-level mode (consistent with other contributed modes)

The implementation is straightforward since you've provided a complete, well-structured mode configuration.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.