agentleFS
tamldtamld/defense-in-depth/.agents/skills/skill-deep-research/SKILL.md

Deep Research

Skill4 starsChanged 7 months agoMITView on GitHub

Structured research workflow for answering complex questions, evaluating technologies, and cross-checking information across multiple sources. Turns curiosity into verified, actionable intelligence.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before searching anything, define what you're looking for and how you'll judge it.

    line 42
  2. Distinctive:

    Start searching before defining what you're looking for

    line 66
  3. Distinctive:

    Search with vague queries ("best framework 2024")

    line 67
  4. Distinctive:

    Search broadly across multiple source categories:

    line 74
  5. Distinctive:

    Search only one category — always include at least 2 categories

    line 114

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

🔬 Deep Research — From Question to Verified Answer

"A question without research is an opinion. Research without cross-checking is propaganda."

This skill transforms questions, ideas, and uncertainties into verified, evidence-tagged intelligence through a structured research pipeline.


Research Pipeline

flowchart TD
    INPUT["Question / Idea /\nUncertainty"] --> P1["① FRAME\nDefine scope & criteria"]
    P1 --> P2["② DIVERGE\nCast wide net"]
    P2 --> P3["③ CROSS-CHECK\nVerify across sources"]
    P3 --> P4["④ EVALUATE\nScore & rank findings"]
    P4 --> P5["⑤ SYNTHESIZE\nRecap & recommend"]
    P5 --> OUTPUT["Verified answer\nwith evidence chain"]
    
    P3 -->|"Contradiction\nfound"| P2

Phase ①: FRAME — Define the Research Scope

Before searching anything, define what you're looking for and how you'll judge it.

Scope Template

**Question:** {The precise question to answer}
**Type:** comparison | evaluation | exploration | fact-check
**Criteria:** {What makes an answer good?}
**Constraints:** {Time, language, recency, domain}
**Anti-Goals:** {What we're NOT trying to learn}

Evaluation Criteria (pick relevant ones)

CriterionDescriptionWeight
RecencyWhen was this published/updated?🔴 High
AuthorityWho published it? (official docs > blog > forum)🔴 High
RelevanceDoes it answer the actual question?🔴 High
AdoptionHow widely used/starred/downloaded?🟡 Medium
ConsensusDo multiple independent sources agree?🟡 Medium
DepthSurface overview vs deep technical detail?🟢 Low

Do Not Do

  • Start searching before defining what you're looking for
  • Search with vague queries ("best framework 2024")
  • Set no criteria — you'll drown in results without a filter

Phase ②: DIVERGE — Cast a Wide Net

Search broadly across multiple source categories:

Source Categories

CategorySourcesTrust Level
OfficialDocs, specs, RFCs, GitHub repos🟢 High
AcademicPapers, conferences, research labs🟢 High
IndustryTech blogs (major cos), benchmark reports🟡 Medium
CommunityStack Overflow, Reddit, HN, Discord🟠 Low-Med
IndividualPersonal blogs, tutorials, tweets🔴 Low

Search Strategy

flowchart TD
    QUERY["Research question"] --> S1["Search 1:\nOfficial docs"]
    QUERY --> S2["Search 2:\nGitHub repos\n(stars, activity)"]
    QUERY --> S3["Search 3:\nTech blogs\n(recent, authoritative)"]
    QUERY --> S4["Search 4:\nCommunity discussions\n(real-world experience)"]
    
    S1 --> POOL["Source pool\n(≥ 5 sources)"]
    S2 --> POOL
    S3 --> POOL
    S4 --> POOL

For Each Source, Record

FieldDescription
URLThe source link
TitleArticle/doc title
Author/OrgWho published it
DateWhen published/updated
CategoryOfficial/Academic/Industry/Community/Individual
Key ClaimsWhat does this source say? (summarize in ≤ 3 points)

Do Not Do

  • Rely on a single source — minimum 5 sources for any research
  • Search only one category — always include at least 2 categories
  • Ignore publication date — a 2020 article about 2024 tech is dangerously outdated
  • Accept the first result — the best answer is rarely the first one

Phase ③: CROSS-CHECK — Verify Across Sources

The most critical phase. No claim is accepted until verified by ≥ 2 independent sources.

Cross-Check Matrix

ClaimSource ASource BSource CVerdict
{Claim 1}✅ Confirms✅ Confirms—Verified
{Claim 2}✅ Confirms❌ Contradicts✅ ConfirmsDisputed — investigate
{Claim 3}✅ Confirms——Unverified — needs more sources

Contradiction Resolution

flowchart TD
    CONTRA["Sources disagree"] --> Q1{"Which source is\nmore authoritative?"}
    Q1 -->|"Official docs"| OFFICIAL["Trust official\nTag: [CODE]"]
    Q1 -->|"Both equal"| Q2{"Which is\nmore recent?"}
    Q2 -->|"Source A newer"| RECENT["Trust recent\nTag: [INFER]"]
    Q2 -->|"Same date"| Q3{"Can you\ntest it?"}
    Q3 -->|"Yes"| TEST["Run experiment\nTag: [RUNTIME]"]
    Q3 -->|"No"| FLAG["Flag as disputed\nTag: [HYPO]"]

Do Not Do

  • Accept a claim from a single source as fact
  • Ignore contradictions — they are the most valuable signals
  • Trust a blog post over official documentation
  • Skip date verification — tech information has a half-life

Phase ④: EVALUATE — Score & Rank

Score each finding against the criteria defined in Phase ①.

Scoring Template

FindingRecencyAuthorityRelevanceAdoptionConsensusTotal
{Finding A}5/54/55/53/54/521/25
{Finding B}3/55/54/55/55/522/25
{Finding C}2/52/53/51/52/510/25

Evidence Level Assignment

Score RangeEvidence LevelMeaning
20-25[RUNTIME] or [CODE]Strong, verified evidence
15-19[INFER]Reasonable confidence, pattern-based
10-14[HYPO]Hypothesis, needs more verification
< 10DropNot trustworthy enough to include

Do Not Do

  • Include low-score findings without flagging their evidence level
  • Let personal preference override scoring data
  • Skip scoring because "the answer is obvious"

Phase ⑤: SYNTHESIZE — Recap & Recommend

Produce a structured research report.

Output Template

## Research Report: {Question}

### Executive Summary
{2-3 sentences answering the original question}

### Methodology
- Sources consulted: {N}
- Source categories: {list}
- Cross-checks performed: {N}

### Key Findings

| # | Finding | Evidence | Sources |
|:---|:---|:---:|:---|
| 1 | {finding} | [RUNTIME] | {source links} |
| 2 | {finding} | [INFER] | {source links} |

### Contradictions & Disputes
{Any unresolved disagreements between sources}

### Recommendation
{What the agent recommends, with reasoning}

### Confidence Level
{HIGH / MEDIUM / LOW — justified by evidence quality}

### Source Registry
| # | Source | Category | Date | Trust |
|:---|:---|:---|:---|:---:|
| 1 | {URL} | Official | 2024-01 | 🟢 |
| 2 | {URL} | Community | 2023-11 | 🟠 |

Do Not Do

  • Present findings without citing sources
  • Give a recommendation without explaining the reasoning
  • Hide low confidence behind assertive language
  • Skip the contradictions section — honesty builds trust
  • Present research as exhaustive when time-boxed — state limitations

Quick Reference

┌────────────────────────────────────┐
│  🔬 Deep Research Pipeline         │
├────────────────────────────────────┤
│  ① FRAME    — Define question      │
│  ② DIVERGE  — ≥5 sources, ≥2 cats │
│  ③ CROSS    — ≥2 sources per claim│
│  ④ EVALUATE — Score & evidence tag │
│  ⑤ SYNTH    — Report + recommend  │
├────────────────────────────────────┤
│  Trust: Docs > Papers > Blogs     │
│  Min sources: 5                   │
│  Cross-check: 2+ per claim        │
│  Evidence: [CODE] [RUNTIME]       │
│            [INFER] [HYPO]         │
└────────────────────────────────────┘

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.