Deep Research

What it does

Produce a comprehensive, cited research report with multi-source evidence tracking, counter-review, and confidence markers.

Procedure

When this skill is activated, Chalie follows these steps:

1. Use memory to recall any prior research, related findings, or user preferences — then ask for the specific question, scope (company vs. industry vs. policy/tech), and preferred output length.
2. Break the topic into 3–6 parallel sub-questions (what, why, how, who, trends, counter-claims) and assign each to a subagent to search independently, keeping raw search context out of the main thread.
3. Each subagent uses search and news to gather sources and read to extract content from the most authoritative URLs, then returns distilled notes — never raw results.
4. Use document to build a citation registry from the distilled notes: record source URL, type (news/official/academic/industry), recency date, and authority level for every claim.
5. Use document to construct an evidence-mapped outline, flagging claims lacking two independent sources and any counter-claims or competing perspectives.
6. Use document to draft the report from the outline and citation registry, embedding confidence markers (confirmed / likely / contested / unverified) per major claim.
7. Use document to run a counter-review pass: for each key finding, state the strongest plausible objection or alternative interpretation and address it inline.
8. Present the final report from document with executive summary, section findings with inline citations, counter-claims addressed, and a confidence summary — then use memory to store key findings.

Version

v1 (curated)