Diglet

❌ WRONG (sequential execution):

WebSearch #1 → wait for results → WebSearch #2 → wait → WebSearch #3...

If you catch yourself doing sequential searches, STOP and redo them in parallel.

MANDATE 2: Source Depth

Minimum unique sources by mode:

• quick: 10+
• standard: 15+
• deep: 25+
• ultradeep: 30+

Thoroughness over speed. Quality > speed.

MANDATE 3: Anti-Hallucination (Zero Tolerance)

• Every factual claim MUST cite a specific source immediately [N]
• Use "According to [1]..." or "[1] reports..." — NEVER "Research suggests..." or "Studies show..."
• Mark inferences as "This suggests..." not stated as fact
• When uncertain: say "No sources found for X" — NEVER fabricate a citation

MANDATE 4: Bibliography Completeness (Zero Tolerance)

Report is UNUSABLE without complete bibliography.

• Include EVERY citation [N] used in report body — all of them, individually
• NO placeholders: never "[8-75] Additional citations"
• NO ranges: write [3], [4], [5]... individually, NOT "[3-50]"
• NO truncation: 30 sources cited = 30 bibliography entries written in full

Report Output Directory — MANDATORY

All reports go to research-reports/ in the project root:

research-reports/[topic-slug]-[YYYY-MM-DD]/
├── report.md              ← Main report
└── sources.md             ← Optional: raw source notes

Directory naming: [topic-slug]-[YYYY-MM-DD] (lowercase, hyphenated) Always create the subdirectory first: mkdir -p research-reports/[topic-slug]-[date]/ Never write to /tmp/ — always use research-reports/ so findings persist. Check existing project files first: Read relevant files in contexts/ or memory/ before searching.

File index update (MANDATORY after saving report): Update docs/file-index.md — add a row under research-reports/ with path and one-line description.

Project linking (if research relates to a project under contexts/):

1. Add projects: [project-slug] to the report's markdown frontmatter
2. Update the project's README.md (add to Research Reports table + Updates Log)

Decision Tree

Request Analysis
├─ Simple lookup? → STOP: Use WebSearch, not this skill
├─ Debugging? → STOP: Use standard tools
└─ Complex analysis needed? → CONTINUE

Mode Selection
├─ Initial exploration? → quick (3 phases, 2-5 min)
├─ Standard research? → standard (6 phases, 5-10 min) [DEFAULT]
├─ Critical decision? → deep (8 phases, 10-20 min)
└─ Comprehensive review? → ultradeep (8+ phases, 20-45 min)

Workflow

1. Clarify (Rarely Needed — Prefer Autonomy)

DEFAULT: Proceed autonomously. Derive assumptions from query signals.

Only ask if CRITICALLY ambiguous:

• Query is incomprehensible (e.g., "research the thing")
• Contradictory requirements

Default assumptions:

• Technical query → technical audience
• Comparison query → balanced perspective
• Trend query → recent 1-2 years
• Standard mode is default

2. Plan

Mode selection:

• Quick (2-5 min): Exploration, broad overview
• Standard (5-10 min): Most use cases [DEFAULT]
• Deep (10-20 min): Important decisions, thorough verification
• UltraDeep (20-45 min): Critical analysis, maximum rigor

Announce: selected mode, estimated time, source target. Proceed without waiting.

3. Act (Phase Execution)

All modes execute:

• Phase 1: SCOPE — Define boundaries (method)
• Phase 3: RETRIEVE — Execute parallel search per MANDATE 1 above. (method)
• Phase 8: PACKAGE — Generate report (load output-formatting and writing-standards for this phase)

Standard/Deep/UltraDeep also execute:

• Phase 2: PLAN — Strategy formulation
• Phase 4: TRIANGULATE — Verify 3+ sources per claim
• Phase 4.5: OUTLINE REFINEMENT — Adapt structure based on evidence (method)
• Phase 5: SYNTHESIZE — Generate novel insights

Deep/UltraDeep also execute:

• Phase 6: CRITIQUE — Red-team analysis
• Phase 7: REFINE — Address gaps

Progressive Context Loading:

• Load methodology sections on-demand per phase
• Load output-formatting only for Phase 8
• Load writing-standards only for Phase 8
• If report exceeds 18K words, load continuation-protocol

Critical: Avoid "Loss in the Middle"

• Place key findings at START and END of sections, not buried
• Structure: Summary → Details → Conclusion

4. Verify

Step 1: Citation Verification

python scripts/verify_citations.py --report [path]

Checks: DOI resolution, title/year matching, flags suspicious entries. If suspicious: review, remove/replace fabricated sources, re-run until clean.

Step 2: Structure & Quality Validation

python scripts/validate_report.py --report [path]

8 checks: executive summary length, required sections, citation format, bibliography matches, no placeholders, word count, minimum sources, no broken links.

If fails: Attempt 1 auto-fix → Attempt 2 manual fix → After 2 failures: STOP, report, ask user.

5. Report

Load output-formatting for full pipeline. Load writing-standards for quality requirements.

Length requirements:

• Quick: 2,000+ words
• Standard: 4,000+ words
• Deep: 6,000+ words
• UltraDeep: 10,000+ words (no upper limit)

Generate three formats: Markdown (primary) + HTML (McKinsey style) + PDF Deliver: Executive summary inline, folder path, format confirmation, source count, next steps.

Output Contract

Required sections (all must be substantive):

• Executive Summary (50-250 words)
• Introduction (question, scope, methodology, assumptions)
• Main Analysis (4-8 findings, each 300-500+ words with citations)
• Synthesis & Insights (patterns, novel insights, implications)
• Limitations & Caveats (gaps, assumptions, uncertainties)
• Recommendations (immediate actions, next steps, further research)
• Bibliography (COMPLETE — see MANDATE 4)
• Methodology Appendix

Quality gates:

• Minimum word count (mode-dependent, see above)
• Average credibility score >60/100
• 3+ sources per major claim
• Clear facts vs. analysis distinction
• All sections present and substantive

Strictly prohibited:

• Placeholder text (TBD, TODO, [citation needed])
• Uncited major claims
• Short summaries instead of detailed analysis
• Vague statements without specific evidence
• Bullet-point-heavy sections (≥80% must be flowing prose)

Error Handling

Stop immediately if:

• 2 validation failures on same error → Pause, report, ask user
• <5 sources after exhaustive search → Report limitation, request direction
• User interrupts/changes scope → Confirm new direction

Graceful degradation:

• 5-10 sources → Note in limitations, proceed with extra verification
• Time constraint → Package partial results, document gaps
• High-priority critique issue → Address immediately

Inputs & When to Use

Required: Research question (string) Optional: Mode, time constraints, required perspectives, output format

Use when: Comprehensive analysis (10+ sources), comparing approaches, state-of-art reviews, market/trend analysis, technical decisions Do NOT use: Simple lookups, debugging, 1-2 search answers

Reference Files (Load On-Demand)

• Methodology — 8-phase pipeline details
• Output Formatting — Report generation, file org, HTML/PDF pipeline
• Writing Standards — Prose quality, citations, anti-hallucination details
• Continuation Protocol — Auto-continuation for reports >18K words
• Report Template — Output structure template