Rapid Research Loop

When not to use this skill

• Do not use when another skill is a better direct match for the task.
• Do not use when the request is outside this skill's scope.

When to use it

• The user needs a structured briefing, comparison table, FAQ, or short memo derived from multiple sources.
• The task explicitly says "research", "investigate", "summarize", or requests citations/links.
• Another agent already gathered raw notes and you must distill them.

Guardrails

1. Always collect at least three independent sources unless the user provided authoritative material.
2. Keep a running source log (see templates/source_log_template.md) before drafting a summary.
3. Mark uncertainty or missing data instead of fabricating details.
4. Prefer deterministic helper scripts for parsing (CSV/JSON) before reasoning about them.

Step-by-step workflow

1. Clarify the question – restate the core decision or comparison criteria in one paragraph.
2. Plan the evidence sprint – list the minimum facts you must confirm; group them by sub-question.
3. Collect evidence – for each fact, gather a citation with date, outlet, and a one-sentence pull quote.
4. Grade sources – label each entry with primary, secondary, or speculative.
5. Synthesize – transform the evidence table into:
   • a 3-bullet executive summary,
   • a risk/unknowns list,
   • and next-step recommendations.
6. Package output – use templates/research_brief_template.md as the outline and include the source log verbatim at the end.

Output template

## TL;DR (3 bullets)
- ...

## Key Findings
1. Finding + cite
2. ...

## Risks / Unknowns
- ...

## Next Best Actions
- ...

## Source Log
| Source | Date | Key Fact | Link | Confidence |
| --- | --- | --- | --- | --- |

Deterministic helpers to pair with

• read_file / search_code for mining local notes or logs.
• analyze_data for CSV-to-table workflows.
• tavily_search or web_page_reader for live web lookups.