agentleFS
DeiblerDeibler/edmund-harness/skills/teams/research-trio/SKILL.md

research-trio

Skill24 starsChanged 2 months agoMITView on GitHub

Three-agent research pipeline — scout finds sources, summarizer distills them, verifier sanity-checks. Use when the user wants a researched answer on a non-trivial topic and you want balanced depth + quality. Call via `spawn_team` with these three roles; tailor the tasks to the specific topic.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Don't use for:

    line 17
  2. Very specific:

    When the team-done event fires, read_team_results(team_id) gives all three results.

    line 46
  3. Distinctive:

    Don't mention "the team" or the roles.

    line 51

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

research-trio

A three-role team for researched answers. Each role has a clear hand-off via the team shared scratch dir.

When to use

  • "Research X and tell me about it" (substantive topics, not quick-lookups).
  • "Compare these N things" where primary sources matter.
  • "What's the current state of Y?"
  • Anything where a single-shot summarize would be too shallow or unverified.

Don't use for:

  • Simple factual lookups → just answer or summarize the URL inline.
  • Media generation or creative writing → this is a research pattern only.

The three roles

scout — finds primary sources. Typical task shape: "Search the web for authoritative info on {topic}. Drop a file at {shared}/scout-sources.json with 8-12 entries: each {url, title, one-line-summary, date, source-type}. Prioritize primary sources (official docs, announcements, papers) over aggregators. Your final text: one-sentence recap plus the path to scout-sources.json."

summarizer — distills scout's findings. Typical task shape: "Read {shared}/scout-sources.json. Pick the 5 most relevant entries for the user's question. For each, open the URL and produce: {url, 3-bullet takeaway, one-line relevance}. Write to {shared}/summaries.md. Your final text: the 5 takeaways, tight."

verifier — sanity-checks summarizer's picks. Typical task shape: "Read {shared}/summaries.md and the source list at {shared}/scout-sources.json. For each of the 5 summaries: flag anything that looks stale (>2 yrs old on a fast-moving topic), contradicts other entries, or over-claims beyond what the source supports. Your final text: which picks are solid, which need caveats, which should be dropped."

Calling it

spawn_team(members: [
  {role: "scout",      task: "<scout task string, topic interpolated>"},
  {role: "summarizer", task: "<summarizer task string, topic interpolated>"},
  {role: "verifier",   task: "<verifier task string, topic interpolated>"}
])

Edmund interpolates the topic into the task strings. There's no separate template engine — the skill's job is to give you the shape; you write the strings.

After completion

When the team-done event fires, read_team_results(team_id) gives all three results. Synthesize into one reply:

  • Lead with a one-line answer to the user's actual question.
  • Back it up with 3-5 of the strongest findings from summarizer.
  • Fold in verifier's flags ("caveat: two of these are from 2023, so take the numbers lightly.").
  • Don't mention "the team" or the roles. The user sees a clean researched answer, not the machinery.

Tradeoffs

  • Three API calls instead of one. Budget ~30-90s wall clock, ~$0.05-0.20 in tokens.
  • Worth it when the answer needs sources + quality gate. Overkill for quick factual lookups.
  • Sequential-feeling even though members run concurrently — scout's output lands before summarizer typically finishes its first tool call, so practical sequencing works out.

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.