Literature Review

When to use

Use this skill when the user supplies multiple sources — papers, abstracts, excerpts, or articles — and needs a structured thematic review rather than a flat summary or per-source analysis. Literature review is thematic by definition: it organizes the source set by concept or argument, not by publication date.

Distinct from research synthesis: literature review groups sources under auto-detected themes (rows) and evaluates each theme across the source set (columns = sources). Research synthesis produces a 7-column per-source matrix. Use literature review when the output is a review-style document intended for a paper, thesis, or survey chapter.

If the user needs a citation-worthy structured document with explicit agreements, disagreements, and gaps framed at peer-review rigor, prefer research-synthesis instead.

Triggers

• User says "Write a literature review on [X]" or "Give me a lit review of these papers" — highest-confidence direct invocation.
• User pastes 3 or more abstracts, excerpts, or papers with no specific ask — proactively offer literature review.
• User says "I'm writing a survey on [X]" or "I need to cover the field of [Y]" — offer literature review as the primary deliverable.
• User says "I need the lit review chapter for [Z]" (thesis or dissertation context) — offer literature review with expanded gap analysis; thesis gap sections are typically longer than paper gap sections.

Instructions

1. Read all provided sources fully before detecting themes.
2. Auto-detect themes from the source set — do not impose themes from outside the sources. A theme is a concept, argument, or research question that at least two sources address directly or implicitly.
3. State the detected theme count and source count in one line at the top of the output (e.g., "Themes detected: 3 (Capacity, Components, Evolution of model). Sources: 3.") so the user can recalibrate scope before reading further.
4. Build the thematic matrix: rows = themes, columns = sources. Each cell describes how that source addresses the row's theme, using terse noun phrases. Write "—" if the source does not address the theme.
5. For each theme, write a thematic synthesis paragraph that compares how sources treat the theme — do not summarize sources in sequence. Surface agreements, tensions, and nuance across sources.
6. Write the ## Research Gaps section: list 2–5 gaps, each with a bold gap name and a short explanatory clause. A gap is a theme that is under-covered across the source set, a methodological weakness shared by most sources, or an important question none of the sources address.
7. Distinguish primary, secondary, and tertiary sources where relevant. Note if a source is a review paper or meta-review (tertiary); do not treat its claims as primary evidence.
8. If source methodology is relevant to the theme, note methodological differences in the matrix cell or in the synthesis paragraph — sources using incompatible designs cannot be directly compared without flagging the mismatch.
9. Apply the writing-profile rule from ## Writing-profile integration to synthesis paragraphs and gap clauses.
10. Apply the BibTeX extension if the user has supplied citation keys or .bib entries.

Output format

Plain GitHub-flavored markdown in the chat.

Header line (required): One line stating theme count, theme names, and source count. Example: **Themes detected:** 3 (Capacity, Components, Evolution). **Sources:** 3.

Thematic matrix block: Standard markdown table — rows = themes, columns = sources. Cells are terse noun phrases. Write "—" for absent coverage.

Thematic synthesis section: One subsection per theme (use **Theme name.** bold inline header or ### Theme name subheading). Each subsection is a prose paragraph comparing how sources treat the theme — not a per-source summary.

Research Gaps section: Headed ## Research Gaps. Bullet list of 2–5 gaps, each formatted as - **Gap name** — short explanatory clause.

BibTeX-aware extension (conditional): If the user supplies citation keys or .bib entries, render each source column header with @key and use [@key] inline citations in synthesis paragraphs. Otherwise use Author (Year) format.

No Obsidian [[wikilinks]]. No JSON or YAML sidecar. No CSV export. Output is portable across Obsidian, Notion, Apple Notes, Logseq, plain-text editors, and academic writing tools.

Quality criteria

1. Themes auto-detected from the source set — not imposed externally. Theme count and source count stated at top.
2. Every matrix cell traces back to a specific source passage — no hallucinated claims.
3. Thematic synthesis produces cross-source comparison, not per-source summary.
4. Primary, secondary, and tertiary sources are distinguished (noted in matrix or synthesis).
5. Methodology assessment present — notes where sources use incompatible designs.
6. ## Research Gaps section lists 2–5 gaps, each with a short explanatory clause.
7. Matrix cells terse; synthesis prose and gap clauses adapt to writing-profile.md.

Anti-patterns

• Chronological-only ordering of sources, ignoring thematic structure — literature review is thematic by definition.
• No cross-source thematic synthesis — each theme row must compare how sources treat the theme, not list their contents in sequence.
• No research gap identification — a literature review without gaps is a summary, not a review.
• Treating review papers as primary sources — meta-reviews of other reviews must be labeled tertiary.
• Citation drift — a claim in the thematic synthesis cannot be traced back to a specific source column.
• No methodology assessment — ignoring that sources may use incompatible designs (empirical vs. theoretical vs. review).
• Listing sources without critical evaluation — descriptive paraphrases only, no judgment on evidence quality or authority.

Example

Input (3 abstracts):

> Miller (1956). "The Magical Number Seven, Plus or Minus Two." Reviews span-of-attention and immediate-memory experiments across modalities. Proposes capacity ≈ 7±2 items. > > Baddeley (2000). "The Episodic Buffer." Theoretical extension of the 1974 Baddeley & Hitch working-memory model, adding a fourth component (episodic buffer) that integrates phonological loop, visuospatial sketchpad, and long-term memory. > > Cowan (2001). "The Magical Number 4 in Short-Term Memory." Multi-paradigm review (visual arrays, running span, auditory recall) arguing active-focus capacity is ~4 items once chunking and rehearsal are controlled.

Output:

# Literature review: Working-memory capacity and architecture

**Themes detected:** 3 (Capacity, Components, Evolution of model). **Sources:** 3.

## Thematic matrix

| Theme | Miller (1956) | Baddeley (2000) | Cowan (2001) |
|-------|---------------|-----------------|--------------|
| Capacity | ~7±2 items across modalities (conflates chunked & unchunked units) | — (capacity not the focus) | ~4 items active focus (chunking/rehearsal controlled) |
| Components | Single undifferentiated short-term store | 4-component model (central executive, phonological loop, visuospatial sketchpad, episodic buffer) | Attentional-focus model, orthogonal to component theory |
| Evolution of model | Foundational (1956 baseline) | Extends Baddeley & Hitch (1974) with episodic buffer for LTM integration | Refines Miller (1956) downward; attentional rather than structural |

## Thematic synthesis

**Capacity.** The field shifted from Miller's 7±2 (observable span) toward Cowan's ~4 (underlying active focus) as chunking and rehearsal were controlled for. These are not contradictory — they operate at different levels of analysis.

**Components.** Baddeley (2000) treats working memory as structurally partitioned (multiple stores + a buffer). Cowan (2001) treats it as attention-limited (a focus window). Miller (1956) predates both debates.

**Evolution.** Baddeley (2000) is cumulative — extends the 1974 model. Cowan (2001) is revisionary — narrows Miller (1956). The field has diverged from a single-store view into two parallel research programs.

## Research Gaps

- **Neuroimaging evidence absent** — no source cites fMRI or EEG evidence for the proposed components or capacity bounds. A 21st-century neuroimaging review would anchor the structural/attentional debate in biological evidence.
- **Individual differences** — capacity estimates are reported as population means; no source addresses variance across age, working-memory disorders, or expertise-driven chunking differences.
- **Cross-modal interaction** — the phonological loop vs. visuospatial sketchpad boundary is treated as given; no source critically examines whether they genuinely dissociate at the neural level.

Writing-profile integration

Four-tier rule based on output section:

• Matrix cells: Always terse — short noun phrases, no prose. context/writing-profile.md is not consulted; cells are structured data, not sentences.
• Theme + source count line (top of output): one-sentence factual statement, profile-neutral.
• Thematic synthesis paragraphs: Full writing-profile consultation applies — tone, sentence-length preferences, and pet peeves from context/writing-profile.md. Synthesis paragraphs typically exceed 100 words, triggering the adapt-to-profile rule. Apply the user's style to cross-source prose, not to matrix rows.
• Research Gaps bullets: Adapt to writing-profile.md for the short explanatory clause after the bold gap name; the gap name itself stays terse.

Example prompts

• "Conduct a literature review on cognitive load theory using the papers in my Literature/ folder."
• "Organize these 5 papers into themes for a literature review section."
• "What gaps exist in the literature on attention and working memory based on my current sources?"
• "I need the lit review chapter for my thesis on [topic] — start with these abstracts."

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jmlozano1990
• Source: jmlozano1990/Cowork-Starter-Kit
• License: MIT
• Homepage: https://github.com/jmlozano1990/Cowork-Starter-Kit/releases/latest

Install and usage instructions live in the source repository linked above.