reference-reading-summarizer

Reference Reading Summarizer

Turn a reference source into a structured, copyright-safe source card. This skill answers: What does this source say?

Do not force every source into project decisions. Use reference-project-synthesizer after a card exists and the project implication matters.

Skill Directory Layout

<installed-skill-dir>/
├── SKILL.md
├── references/
│   ├── model-routing.md
│   └── reading-modes.md
└── templates/
    ├── source-card.md
    ├── paper-card.md
    └── reading-run.md

Core Contract

• Read to the requested purpose, not exhaustively by default.
• Use cheaper models for skim and routine extraction; escalate for central method, theory, benchmark, closest-work, final-paper claims, or collaborator decisions.
• Preserve provenance: source path, source type, pages/sections/files inspected, reading mode, model tier, confidence, and reviewer status.
• Do not paste long raw source text into cards. Summarize and point to pages/sections/files.
• Separate source content from project implications.

Source Types

• paper-or-pdf: paper PDFs, reports, slide exports, or other PDF references
• collaborator-doc: PDF/Word/text feedback or collaborator-authored notes
• markdown / text-note / latex-source: notes, specs, drafts, or design documents
• bibliography: BibTeX collections and citation sets
• script / notebook / config-or-spec: implementation hints, protocols, task definitions
• bundle: folder containing multiple docs/scripts/bib/configs that should be interpreted as a unit

Reading Modes

• skim: relevance, role labels, whether deeper reading is needed
• extract-writing: intro framing, paragraph moves, contribution wording, captions, figure/table narration
• extract-method: algorithm, objective, architecture, inference, implementation details
• extract-theory: assumptions, theorem statements, proof ideas, formal definitions
• extract-benchmark: task, dataset, split, metric, protocol, compute, evaluation caveats
• extract-baseline: baseline method role, fairness conditions, comparison requirements
• extract-risk: closest-work threat, novelty boundary, reviewer attack surface
• extract-feedback: collaborator comments, requested edits, contradictions, decisions, TODOs
• extract-spec: requirements, constraints, interfaces, acceptance criteria, project assumptions
• extract-bundle: bundle inventory, internal relationships, most useful files, missing context
• extract-implementation-hints: reusable scripts, configs, APIs, preprocessing, commands, pitfalls
• extract-project-seed: initial idea, problem, assumptions, resources, open questions, first actions
• deep-read: high-value source where misunderstanding would change project direction

Model Routing

Read references/model-routing.md.

Default:

• Tier 0 deterministic tools: metadata, file inventory, page/text extraction, bundle manifest
• Tier 1 cheap sidecar: skim, simple card skeleton, writing-pattern extraction, bundle inventory
• Tier 2 normal main model: benchmark/protocol extraction, method extraction, collaborator feedback structuring, citation support
• Tier 3 strong/deep model: core theory, closest work, must-have baseline, final paper claims, project seed, collaborator decision conflict

Escalate when the source is closest work, a core algorithm source, a theory source, a benchmark source, a project seed, a collaborator constraint, or when the card will support a claim, experiment, baseline, implementation, paper revision, or rebuttal.

Workflow

1. Locate reference/.agent/source-index.md first; fall back to reference/.agent/reference-index.md.
2. Identify the source path, source type, and requested reading mode.
3. Create a run artifact under reference/.agent/runs/<run-id>/ when extraction is nontrivial:
   • input-manifest.md
   • extraction-notes.md
   • model.json
   • decision.md
4. Extract only the needed text/pages/files. Prefer local tools; if unavailable, proceed from available metadata and mark limitations.
5. Fill templates/source-card.md into reference/cards/<source-id>.md. For pure papers, templates/paper-card.md remains a compatible subtype.
6. Update reference/.agent/processing-status.md from unread or skimmed to carded when appropriate; update reading-status.md if present for compatibility. Memory writeback: this skill writes only to reference/cards/ and reference/.agent/; project-level memory writeback (claim/evidence/risk boards) is the job of reference-project-synthesizer after synthesis.
7. Route:
   • project implications needed -> reference-project-synthesizer
   • broad field map needed -> literature-review-sprint
   • citation coverage needed -> citation-coverage-audit

Source Card Rules

Every card should include:

• metadata and source type
• reading mode
• role labels
• summary or extracted content only as relevant
• limitations and uncertainty
• reusable assets
• provenance and confidence

Do not turn cheap skim cards into strong conclusions. Mark confidence: skim or needs-deep-read when appropriate.