Skip to content
AI & Agents
Skill

/research-summarizer

Use when the user asks for a sourced briefing on a topic that spans multiple web sources and requires citations.

BOOST
From plugin
llm-agents-ecosystem-handbook
55510 skills
Install
$ npx -y skills add oxbshw/LLM-Agents-Ecosystem-Handbook --skill research-summarizer --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/research-summarizer

Context preview

The summary Claude sees to decide when to auto-load this skill.

Use when the user asks for a sourced briefing on a topic that spans multiple web sources and requires citations.

SKILL.md

research-summarizer.SKILL.md
name: research-summarizer
description: Use when the user asks for a sourced briefing on a topic that spans multiple web sources and requires citations.
version: 0.1.0

Research Summarizer

When to use

  • User asks "research X" / "what's the state of Y" / "summarize the literature on Z"
  • The answer requires multiple sources, not one
  • The user expects citations

When NOT to use

  • Single-source summarization (use a plain summarize tool)
  • Code search (use `repo-auditor` or grep)
  • Opinion / recommendation requests (this skill produces neutral briefings)

Inputs

| Name | Type | Required | Notes | |---|---|---|---| | `topic` | string | yes | the question to research | | `depth` | "shallow" \| "deep" | no | default "shallow" (5–7 sources); "deep" → 10–15 | | `audience` | string | no | shapes vocabulary level (e.g., "executive", "engineer") |

Outputs

A Markdown file `briefing.md` following the structure in `references/report-template.md`.

Workflow

1. **Plan**: produce 3–7 sub-questions covering breadth and depth 2. **Search + fetch**: for each sub-question, run `web_search` then `fetch_url` for top 2–3 hits 3. **Cluster**: group findings by claim; load `references/clustering-rules.md` 4. **Draft**: fill `references/report-template.md` — every claim needs a `[n]` citation 5. **Validate**: every citation resolves; no domain dominates (≥ 3 distinct domains for shallow, ≥ 5 for deep) 6. **Self-review** against success criteria below

References

  • [`references/report-template.md`](references/report-template.md) — the output structure
  • (add `clustering-rules.md`, `domain-allowlist.md` as the skill matures)

Success criteria

  • ≥ 5 sources for shallow, ≥ 10 for deep
  • ≥ 3 distinct domains
  • 0 unsourced claims (or all flagged `(unsourced)`)
  • Open Questions section populated

Failure modes

  • Fewer sources available than required → return partial result, flag at top
  • All sources from one domain → reject, retry plan with broader scope
  • Network error → fail loudly, don't fabricate

Examples

  • See `references/examples/` (TBD) for one shallow and one deep run
Read more
Ships withllm-agents-ecosystem-handbook

A practical operating manual for building, evaluating, securing, and shipping modern LLM agent systems. Modern agents are not "a prompt + a tool."

Get the whole plugin