research-readwise-obsidian

name	research-readwise-obsidian
description	Use the Readwise CLI to run relevance-first research across Reader documents and highlights, then synthesize an Obsidian-ready note with source-linked evidence and explicit confidence gaps. Use when researching topics from Readwise content for PKM workflows in Obsidian.

Use the `readwise` CLI to run deep research on a topic by searching both Reader documents and Readwise highlights, then synthesize an Obsidian-ready research note for PKM.

Scope for this skill is intentionally narrow: search, retrieval, evidence ranking, and synthesis. It does not perform inbox triage, tagging or metadata edits, exports, or archive maintenance unless the user explicitly overrides scope.

<quick_start>

1. Run preflight and auth checks:

command -v readwise
readwise --help

If authentication is missing, run readwise login or readwise login-with-token <token>.

2. Run both searches for the same query in machine-readable mode:

readwise --json reader-search-documents --query "$TOPIC" --limit 20
readwise --json readwise-search-highlights --vector-search-term "$TOPIC" --limit 30

3. Pull deeper context for top documents:

readwise --json reader-get-document-details --document-id "<document_id>"
readwise --json reader-get-document-highlights --document-id "<document_id>"

4. Produce one Obsidian note using the fixed template in <process>. </quick_start>

1. Normalize the input topic into: - one primary research question - 3 to 6 subthemes for sectioning - 3 to 8 query variants (synonyms, alternate phrasings, opposing terms)

2. Execute initial retrieval across both systems:

• Reader documents: reader-search-documents
• Readwise highlights: readwise-search-highlights Use relevance-focused limits first (documents 20, highlights 30). Increase only when recall is weak.

3. Rank evidence by relevance before recency. Score each candidate snippet with this rubric:

• 3: directly answers the research question with specific claim or detail
• 2: supports a subtheme with useful context
• 1: tangential but potentially useful
• 0: irrelevant or noise Keep score 2-3 items first. Use recency only as a tiebreaker.

4. Enrich top evidence with source context. For top-ranked Reader documents, fetch:

• full markdown and details (reader-get-document-details)
• document highlights (reader-get-document-highlights) Prefer exact highlighted text over paraphrase whenever available.

5. Run refinement loops only if evidence is weak. Weak evidence means any subtheme has fewer than 2 strong snippets (score >= 2) or conflicting claims without sufficient support. Perform up to 2 refinement rounds:

• round strategy A: broaden terms (synonyms, adjacent concepts)
• round strategy B: narrow terms (specific entities, date constraints) Stop early once evidence quality is sufficient.

6. Enforce section-level snippet caps. For each subtheme section, include at most 5 snippets. Select highest relevance first, then maximize source diversity.

7. Resolve conflicting evidence explicitly. When sources disagree:

• present both views
• attach source-linked evidence for each view
• explain likely reasons for disagreement (timeframe, domain, methodology, or context)

8. Produce the final fixed Obsidian note. Use this structure exactly:

---
title: "{{YYYY-MM-DD topic-slug}}"
created: "{{ISO-8601 datetime}}"
topic: "{{research topic}}"
source_system: "readwise-cli"
tags:
  - research
  - readwise
  - obsidian
  - pkm
---

# {{research topic}}

## Research Question
{{one-sentence question}}

## Key Findings
- {{finding 1}}
- {{finding 2}}
- {{finding 3}}

## Evidence by Theme
### {{theme 1}}
1. "{{verbatim snippet}}" — [{{source title}}]({{source url}}) _(author: {{author|unknown}}, category: {{category|unknown}}, document_id: {{document_id|n/a}}, highlight_id: {{highlight_id|n/a}})_
2. "{{verbatim snippet}}" — [{{source title}}]({{source url}}) _(author: {{author|unknown}}, category: {{category|unknown}}, document_id: {{document_id|n/a}}, highlight_id: {{highlight_id|n/a}})_

### {{theme 2}}
1. "{{verbatim snippet}}" — [{{source title}}]({{source url}}) _(author: {{author|unknown}}, category: {{category|unknown}}, document_id: {{document_id|n/a}}, highlight_id: {{highlight_id|n/a}})_

## Conflicting Evidence
- **View A:** {{claim summary}}
  Evidence: [{{source title}}]({{source url}}) _(author: {{author|unknown}}, category: {{category|unknown}}, document_id: {{document_id|n/a}}, highlight_id: {{highlight_id|n/a}})_
- **View B:** {{claim summary}}
  Evidence: [{{source title}}]({{source url}}) _(author: {{author|unknown}}, category: {{category|unknown}}, document_id: {{document_id|n/a}}, highlight_id: {{highlight_id|n/a}})_
- **Why they differ:** {{reason grounded in source context}}

## Confidence and Gaps
  {{High|Medium|Low}} — {{justification}}
  {{missing or underrepresented evidence}}
 
 {{query refinement 1}}
 {{query refinement 2}}


 []() — author: {{author|unknown}}, category: {{category|unknown}}, location: {{location|n/a}}, documentid|n/a}}, highlightid|n/a}}
 []() — author: {{author|unknown}}, category: {{category|unknown}}, location: {{location|n/a}}, documentid|n/a}}, highlightid|n/a}}

9. Citation requirements are mandatory:

• citations must be markdown links to source URLs
• each citation must include Readwise metadata inline (author, category, document_id, highlight_id when available)
• if a URL is unavailable, explicitly mark url unavailable and keep metadata

10. Obsidian naming guidance:

• recommended note filename format: YYYY-MM-DD topic-slug.md
• keep slugs lowercase with hyphens
• keep note title aligned with filename stem

<success_criteria> This skill is successful when all conditions are true:

• Searches both Reader documents and Readwise highlights for the topic.
• Uses relevance-first ranking and only uses recency as secondary ordering.
• Runs at most 2 query refinement rounds, and only when evidence is weak.
• Limits evidence to a maximum of 5 snippets per theme section.
• Produces the fixed Obsidian note structure without omitting sections.
• Includes a Conflicting Evidence section with both views when disagreement exists.
• Always includes a Confidence and Gaps section, even when confidence is high.
• Uses markdown-link citations with Readwise metadata for every evidence item.
• Stays within scope: search, retrieval, and synthesis only unless user explicitly requests broader workflows. </success_criteria>