AgentSkillsCN

literature-review

文献综述

SKILL.md

Literature Reviewer Skill

Invocation

code
/literature-review [phase]

Where [phase] is one of:

  • •required-reading - Process professor-provided texts
  • •explore - Exploratory searches to discover the field
  • •define-domains - Define domains based on exploration
  • •search [domain] - Systematic search for a specific domain
  • •critique [domain] - Run critique loop on domain summary
  • •synthesize - Write unified review chapter
  • •status - Show current progress

Workflow Overview

Phase 1: Required Reading

Input: 7 professor-provided texts in academic/texts/ Output: Understanding of what Artistic Research is

For each text:

  1. •Read and extract key concepts
  2. •Add to ../submission/references/bibliography.bib
  3. •Document: What does this tell me about AR?
  4. •Document: How does this relate to Kepler project?
  5. •Update checkpoint.md

Phase 2: Exploration

Input: Search queries from CLAUDE.md Output: Understanding of the AR field, citation landscape

Steps:

  1. •Run exploratory searches across venues
  2. •Document findings (topics, citation counts, gaps)
  3. •Identify emerging themes/clusters
  4. •Note typical citation counts in AR field
  5. •Update checkpoint.md with findings

Phase 3: Domain Definition

Input: Exploration findings Output: Domain definitions, calibrated tier thresholds

Steps:

  1. •Review exploration notes
  2. •Identify 3-5 coherent domains
  3. •Calibrate citation tier thresholds for AR field
  4. •Update CLAUDE.md with domain definitions
  5. •Update todo.md with domain-specific tasks

Phase 4: Systematic Search (per domain)

Input: Domain definition, search terms Output: Domain summary, BibTeX entries

Steps:

  1. •Load context from checkpoint.md
  2. •Search using MCPs (Semantic Scholar, OpenAlex)
  3. •Search AR venues (JAR, PARSE, VIS, RC)
  4. •Triage papers by tier
  5. •Write domain summary (1500-2500 words)
  6. •Export BibTeX to data/exports/domain_*.bib
  7. •Update checkpoint.md

Phase 5: Critique Loop

Input: Domain summary draft Output: Revised summary, critique log

Steps:

  1. •Grade summary A-F on:
    • •Completeness (all tiers covered?)
    • •Coherence (logical flow?)
    • •Relevance (connects to Kepler project?)
    • •Citation quality (tier distribution appropriate?)
  2. •Log critique to reviews.log
  3. •Revise based on feedback
  4. •Re-grade until B+ or better

Phase 6: Synthesis

Input: All domain summaries Output: Unified review chapter

Steps:

  1. •Read all domain summaries
  2. •Write unified systematic_review_chapter.md
  3. •Update ../submission/docs/02-literaturrecherche.md
  4. •Merge BibTeX into ../submission/references/bibliography.bib
  5. •Build PDF: cd ../submission && make literatur
  6. •Final critique loop on unified chapter

Critique Grading Rubric

GradeCriteria
AComprehensive, well-structured, all tiers covered, clear relevance
BGood coverage, minor gaps, coherent structure
CAdequate but missing key papers or weak connections
DSignificant gaps, poor structure, unclear relevance
FIncomplete, major issues, requires substantial revision

Minimum passing grade: B

File Outputs

FileContent
checkpoint.mdSession state after each phase
todo.mdUpdated task list
reviews.logAll critique feedback
data/exports/*.bibBibTeX per domain
drafts/*.mdDomain summaries

Integration Points

Final outputs go to:

  • •../submission/docs/02-literaturrecherche.md - Chapter content
  • •../submission/references/bibliography.bib - Citations

Build command:

bash
cd ../submission && make literatur

Session Recovery

If context fills or session restarts:

  1. •Read checkpoint.md first
  2. •Resume from last incomplete phase/domain
  3. •Continue workflow from that point