deep-research
npx machina-cli add skill arudita-zzz/app-dev-prompt-suite/deep-research --openclawYou are a research orchestrator. You decompose complex research topics into focused investigation tasks, execute them serially via Task tool subagents, adaptively refine the plan after each task, and synthesize findings into a comprehensive report.
Steps
0. Init
- Read conventions for defaults
- If argument is provided, use it as the research topic. Otherwise, AskUserQuestion for the research topic.
- AskUserQuestion for
research_name(used for directory naming, short kebab-case identifier) - Set output path:
{docs_dir}/{research_name}/research/ - Create subdirectory:
{docs_dir}/{research_name}/research/tasks/ - TaskCreate:
Deep Research: <research_name>
1. Task Decomposition
Read task decomposition instructions and execute.
2. Investigation Loop
Read investigation loop instructions and execute.
3. Synthesis
Read synthesis instructions and execute.
4. Report
- Use document-summarizer agent to create max 80-line summary of
research_report.md - Display the summary to the terminal
- AskUserQuestion:
- Approve report
- Request revisions
- Request additional investigation on specific topics
- Type Anything
5. Complete
- TaskUpdate: mark completed
- Display report path:
{docs_dir}/{research_name}/research/research_report.md - AskUserQuestion: proceed to feasibility-study with this research?
- Yes → display:
/enterprise-dev-suite:feasibility-study -r {path-to-research_report.md} - No → end
- Type Anything
- Yes → display:
Constraints
- Document language: see conventions
- Docs dir:
.claude/claudeRes/docs(.claude is project-level one) - All investigation artifacts are saved under the research output path
- Investigation loop runs fully autonomously — no user interaction between Step 1 approval and Step 4 report
Source
git clone https://github.com/arudita-zzz/app-dev-prompt-suite/blob/main/plugins/enterprise-dev-suite/skills/deep-research/SKILL.mdView on GitHub Overview
Deep-research breaks a complex topic into focused investigation tasks, executes them serially via Task tool subagents, and adapts the plan after each task. It synthesizes findings into a comprehensive report, stored in a structured project directory. This approach enables thorough, autonomous exploration with minimal manual intervention.
How This Skill Works
It starts from a research topic or question. The system decomposes it into investigation tasks, runs them serially via Task subagents, and adaptively refines the plan based on results. Finally, findings are synthesized into a comprehensive report using a document-summarizer to create an 80-line summary.
When to Use It
- When breaking down a complex research topic into focused investigation tasks is needed
- When you want autonomous task execution with adaptive re-evaluation between steps
- When generating a comprehensive, synthesis-backed research report is required
- When you want all artifacts stored under the designated docs_dir with a kebab-case name
- When conducting web-enabled investigations to gather diverse sources (WebFetch, WebSearch)
Quick Start
- Step 1: Provide the research topic (argument) and a kebab-case research_name to initialize the workspace
- Step 2: Let the Task subagents decompose tasks and execute investigations autonomously, refining the plan after each task
- Step 3: Generate the research_report.md and run the document-summarizer to produce an 80-line summary for review
Best Practices
- Define a precise research topic and a kebab-case research_name before starting
- Let the investigation loop run autonomously between Task 1 and the final report
- Keep all artifacts under the designated docs_dir and research-specific folder
- Use the document-summarizer to generate an 80-line summary of the final report
- Review the final research_report.md and decide on pursuing a feasibility study if needed
Example Use Cases
- Literature review on neural network generalization and transfer learning
- Market landscape analysis for a new AI-enabled healthcare product
- Regulatory impact assessment for data privacy and security standards
- Competitive analysis of cloud-native observability tools
- Technical investigation into energy efficiency for edge computing devices
