agentleFS
fortunto2fortunto2/solo-factory/skills/deep-research/SKILL.md

solo-deep-research

Skill18 starsChanged 4 months agoMITView on GitHub

Deep research on any question from three independent angles in parallel, every claim tied to a source, contradictions checked before the answer. Use when "deep research", "дип ресерч", "глубокое исследование", "разберись глубоко", "parallel research", "research it from several angles", or a decision rests on facts one quick search will not settle. Do NOT use for a quick idea scan (use /research), idea scoring and PRD (use /validate), or handing coding tasks to worker agents (that is the swarm tool, not research).

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use it when being wrong is expensive: a market to enter, a tool to bet on, a legal or financial fact, a technical choice.

    line 19
  2. Very specific:

    What is already known: If MCP kb_search is available, search the local knowledge base first.

    line 24
  3. Distinctive:

    The question, sharpened: Write the one question the answer must settle and what would change the decision.

    line 27
  4. Distinctive:

    Prefer primary sources: the vendor's docs and pricing page, the filing, the paper, the statute, the repo.

    line 45
  5. Distinctive:

    Not found is a finding: say what was searched and that it came back empty.

    line 48
  6. Distinctive:

    No source, no claim. A guess is labelled as a guess.

    line 49
  7. Distinctive:

    Cause: only the built-in search is available.

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/deep-research

Research "$ARGUMENTS" deeply: three angles in parallel, each claim with its source, then one pass that checks the claims against each other before anything is concluded.

Use it when being wrong is expensive: a market to enter, a tool to bet on, a legal or financial fact, a technical choice. For a first look at an idea, /research is cheaper.

1. Before searching

  • What is already known. If MCP kb_search is available, search the local knowledge base first; if session_search is available, check whether this was researched before. Start from what exists and say what it already answers.
  • The question, sharpened. Write the one question the answer must settle and what would change the decision. A vague topic becomes 3–5 concrete sub-questions.
  • Three angles that do not overlap. Pick them for this question. For a startup idea the default is market (competitors, size, pricing), users (pain, sentiment, quotes) and tech (feasibility, open-source, effort). For a tool: capability, cost and limits, what users report. For a fact: primary sources, secondary reporting, counter-evidence.

2. Three researchers in parallel

Spawn three subagents (Agent tool) in one message so they run at once. Each brief carries the question, its own angle, what is already known, and the source rules below. Each returns a list of findings, every one with a URL and the date of the source.

Agent teams work too when available (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1), but plain subagents are enough: the researchers do not need to talk to each other, the lead compares them.

Source rules for every researcher:

  • Each claim carries a link to where it was read, and the source's date.
  • Prefer primary sources: the vendor's docs and pricing page, the filing, the paper, the statute, the repo. A blog quoting them is a pointer to the primary, not a source.
  • Numbers keep their unit, date and scope ("$4.2B global TAM 2025, per X", not "a big market").
  • Not found is a finding: say what was searched and that it came back empty.
  • No source, no claim. A guess is labelled as a guess.

Search backends, in this order where available: MCP web_search (engine routing, site: filters), WebSearch (built in), MCP web_extract or WebFetch to read a page in full. Targeted: site:reddit.com, site:news.ycombinator.com, site:github.com, official docs.

3. Check before concluding

The lead, not a researcher:

  • Lines up the three lists: where they agree, where they contradict, what only one saw.
  • Opens the source of every claim the conclusion rests on and confirms it says that, with the date. A claim whose source does not hold it is dropped and named as dropped.
  • Resolves contradictions by the better source (primary over secondary, newer over older) or leaves them open and says so.
  • Separates what is verified from what is reported and from what is inferred.

4. Output

docs/research.md in the current project (for an idea, the format from /research), or the answer in chat when there is no project. It holds:

  1. Answer first: the question and the conclusion in two or three sentences, with how sure.
  2. Findings by angle, each with its link and date.
  3. Contradictions and open questions, and what would settle them.
  4. Verdict: for an idea GO / NO-GO / PIVOT and the next step (/validate <idea>); for other questions, the recommendation and what it depends on.

Common issues

The three angles return the same links

Cause: the angles overlap or the briefs are generic. Fix: make each brief name what the angle must NOT cover, and give each its own search terms.

Web search returns little

Cause: only the built-in search is available. Fix: WebSearch still works. For richer results with engine routing (Reddit, GitHub, YouTube), set up SearXNG (self-hosted, free) and the solograph MCP.

A finding cannot be traced back

Cause: a researcher summarised without keeping the URL. Fix: the claim does not go into the answer. Ask that researcher for the source or drop it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.