deep-research-framework – skill Dữ liệu & Khoa học cho agent AI

Deep Research Report Framework

Produce a report whose "conclusions are trustworthy, process is traceable, and the unknowns are explicitly marked." The quality floor of research is set by its sources, not its prose.

When to Use

• Use when the user asks for a research report, topic investigation, or competitive analysis.
• Use when sources need tiering and cross-verification before writing conclusions.

Workflow

1. Define the question: write in one sentence what the report must answer. Add at most 3 sub-questions. Starting before the question is clear guarantees the research will sprawl.

2. Tier your sources (tier before you search, so you don't believe whatever you find first):

• Primary: official docs, original papers, original filings, protocol/legal texts, measured data.
• Secondary: reputable press, industry reports, expert interviews.
• Tertiary: aggregator summaries and unsourced commentary — leads requiring corroboration. A direct author statement, original release or first-hand report on a forum/social platform may be primary evidence for that narrow claim; assess provenance and corroboration rather than classifying by hosting platform alone.

3. Cross-verify: key conclusions need two or more independent sources. Two sources quoting the same origin don't count as independent. If no second source exists, downgrade the conclusion to "single-source claim" and mark it as such.

4. Writing structure (conclusion first):

• Opening: 3–5 lines stating the core conclusions.
• Body: one section per sub-question, each in "conclusion → evidence → sources" order.
• Closing: uncertainty statement — "what wasn't found, what's speculation, under what conditions the conclusions change."

5. Timestamps: date every key fact ("as of Sep 2026"). Undated information is treated as "possibly stale" by default.

Rules

• Mark speculation as speculation. Use "may", "likely", "unverified" — never dress speculation up with "clearly" or "it is well known".
• Do not base a core conclusion on unverified commentary. Assess each item’s provenance and what it directly establishes; a genuine primary protocol proposal hosted on a forum can establish the proposal, but not deployment or correctness.
• Every number in the report has a source. A number without a source is deleted, not rewritten.
• State "what you don't know" before "what you know" — honest uncertainty beats pretty certainty.

Minimal example (report outline template)

# Research Report: <Topic> (example outline)

**Core conclusions** (3–5 lines): …

## 1. Sub-question one
- Conclusion: …
- Evidence: … (source: primary/secondary, as of …)
- Evidence: … (second independent source)

## 2. Sub-question two
…

## Uncertainty statement
- Not found: …
- Single-source claims: … (one source only, pending verification)
- Invalidation conditions: if … changes, conclusion X in this report needs reassessment.

Limitations

• Source quality is the ceiling of the report. Declared "primary sources" cannot be verified for authenticity by this skill; it only enforces the tiering and cross-verification discipline.
• Independent-source requirements are a judgment call. Two outlets quoting the same wire story or press release count as one origin, and the skill cannot detect shared-origin syndication automatically.
• Web access is required for source-backed research. Without search or fetch tools the skill stops rather than writing a report from memory, or marks every external fact [UNVERIFIED] and warns on the cover.
• Recency claims expire. A dated fact ("as of Sep 2026") is only valid until the underlying source changes; the skill does not re-verify older reports.

Anti-patterns

• ❌ Single-source verdicts: one social-media post becomes "research shows".
• ❌ Speculation as fact: "clearly" and "it is well known" are the most dangerous words in a research report.
• ❌ Undated facts: 2023 data presented as the current state; the conclusion quietly expires.
• ❌ Evidence pile with no conclusion: 20 sources listed, and the reader still can't tell what the author believes.
• ❌ Hiding the unknowns: burying what wasn't found makes the report look omniscient and plants landmines.