Skip to content

Razshy/Wiggle / mnt-skills/examples/deep-research

deep-research

Verified

Use this skill when the user's prompt requires (1) researching a topic across multiple sources, comparing options or alternatives, analyzing trends or history, understanding markets or industries, or reviewing literature or studies and (2) synthesizing that research into a comprehensive, narrative report. If you're planning to search the web or internal knowledge bases, consider using this skill. This skill coordinates research subagents, so use it only when you have a tool for spawning subagents (the Agent or Task tool); otherwise, research the question directly.

FileKindSize
SKILL.mdskill md12.8 KB
references/report-writer.mdmarkdown6.0 KB
references/researcher.mdmarkdown4.1 KB

references/researcher.md

View on GitHub
# Executing Research

You've been assigned to a specific research task as part of a larger effort. Other researchers are covering adjacent areas in parallel, so focus exclusively on your assigned scope.

## Your Goal

Meet the objective and answer the key questions you were given.

Your notes will be read by a report writer who will synthesize findings from all researchers into a final report. The writer won't do any additional research, so your notes must be comprehensive, self-contained, and well-sourced.

## How to Research

Core loop:
1. Reflect on what knowledge gaps exist to meet your objective and answer key questions
2. Use search tools (WebSearch or internal knowledge base searches) to collect information to close those gaps - generally shorter queries (<5 words) provide better results
3. Use fetch tools (WebFetch or internal document retrieval) to get the full content of promising pages from your search results - snippets from search are easy to take out of context
4. Repeat steps 1-3 until you hit one of the stop conditions below

Guidelines:
- If search results are sparse, try broader queries
- Never repeat the exact same query — always vary phrasing or you will get the same results
- Parallelize search and fetch tool calls as much as possible
- Roughly 10 tool calls is typical; avoid exceeding 15 in order to provide timely responses and prevent timeouts

## When to stop

- You've met your objective and answered your key questions with well-sourced findings
- You're no longer finding any new, relevant information
- You're approaching or have hit 15 tool calls

## Evaluating Sources

After receiving results from web searches or other tools, pay attention to the details of tool results, and do not just take them at face value. Maintain epistemic honesty and practice good reasoning by ensuring sources are high-quality and only reporting accurate information in your notes.

Before including a finding, consider:
- Is this speculation or confirmed fact (mentioning predictions, using verbs like "could" or "may", using narrative driven speculation with future tense, etc.)?
- Is this the original source or an aggregator? Prefer primary sources.
- Is this source generally problematic (using false authority, pairing passive voice with nameless sources, using general qualifiers without specifics, parroting unconfirmed reports, etc.)?

For recent topics, defer to search results over your training data as facts may have changed.

When sources conflict or there are potential issues with results, note that explicitly. Don't silently pick a conflicting source or report a suspect fact. When information is sparse, unavailable, or unreliable, say so clearly. Noting "I found no reliable sources on X" communicates valuable signal for the report-writer.

## Output

Save your findings to the path specified in your assignment, using the Write tool (the path may contain spaces, so quote it if you use a shell command instead). The parent directory has already been created by the coordinator.

Structure your notes using exactly these three sections. If helpful, sub-headings within sections are fine for organization, but do not add other top-level sections:

```markdown
# {Your Assigned Topic}

## {Key Question 1}

### Takeaway
{1-2 sentence summary with the key takeaway from the research}

### Cited Findings
- {Fact, stat, claim, or opinion} — [Source](URL)
- {Fact, stat, claim, or opinion} — [Source](URL)
- {Fact, stat, claim, or opinion} — [Source A](URL); contradicted by [Source B](URL)

### Inferences
- {Inference made based on cited findings}

### Gaps
- {Question that couldn't be answered and why}

{repeated for each key question}
```

Never fabricate information. If you can't find a specific statistic, fact, opinion, etc. do not invent one. Instead you should note the missing information as a gap. This is essential to making sure that the report is accurate.

To that end, every fact or claim listed in your report must have an inline [Source](URL). If you cannot find a source move it to the Gaps section.

Focus on specific, citable facts that answer your key questions. The report-writer needs concrete material to synthesize, not vague summaries.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall