Research Brief Playbook

Use this Skill when an agent should move from raw sources to a concise research brief with evidence.

When to use this Skill

Use it when:

• the user wants a brief grounded in public sources
• there may be multiple source pages or supporting documents
• the agent should gather evidence first and summarize second

Do not use it for speculative brainstorming without source collection.

Workflow

1. Start with the smallest source set that can answer the question.
2. If a single page is enough, use a direct page extraction path first.
3. If multiple related pages matter, use a bounded crawl.
4. Convert documents and extract tables only when the brief needs them.
5. Summarize only after the useful evidence is already assembled.

Output standard

Every brief should include:

• the main answer
• the most relevant evidence
• source caveats or missing context
• a short list of follow-up questions when the evidence is incomplete

Use references/research-brief-template.md as the default format. Use references/tool-selection-guide.md when deciding which tool should handle the next step.

Suggested helper paths

Fetch one page:

./scripts/fetch-page.sh '{"url":"https://example.com/report","format":"markdown"}'

Crawl a small site section:

./scripts/crawl-site.sh '{"url":"https://example.com/research","max_pages":5}'

Summarize assembled notes:

./scripts/summarize-brief.sh '{"text":"<assembled research notes>","style":"concise brief"}'

Each helper also accepts:

• --input-file payload.json
• no args to fall through to the underlying agentpm run command

Failure cases to avoid

• Summarizing before checking whether the source is authoritative
• Mixing extracted facts with assumptions
• Treating one vendor blog post as conclusive evidence
• Omitting the strongest counterexample or caveat