Literature Review Sprint

• a user needs to understand a new field quickly
• novelty depends on whether close prior work already exists
• a project needs canonical, closest, and recent/concurrent papers before algorithm or experiment design
• a paper draft has weak related-work positioning but the goal is still field understanding, not final citation cleanup
• an advisor meeting needs a crisp paper map, gap analysis, or next reading plan
• early experiments or writing reveal that the project may be in the wrong literature frame

citation-audit

citation-coverage-audit

• research-project-memory

  when literature findings should persist as risks, actions, claims, or positioning decisions

• research-idea-validator

  before or after the sprint when the result should become a pursue/revise/park/kill decision

• algorithm-design-planner

  when the map clarifies the closest baseline and the method now needs specification

• experiment-design-planner

  when the map implies required baselines, datasets, metrics, or diagnostics

• paper-evidence-board

  when literature risks should be linked to paper claims, sections, figures, and reviewer risks

• citation-coverage-audit

  later, after the paper is close to submission

Skill Directory Layout

<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── memory-writeback.md
    ├── paper-taxonomy.md
    ├── reading-priority.md
    ├── search-protocol.md
    └── synthesis-template.md

Progressive Loading

• Always read

  references/search-protocol.md

  ,

  references/paper-taxonomy.md

  , and

  references/reading-priority.md

  .
• Read

  references/synthesis-template.md

  before writing the final sprint report.
• Read

  references/memory-writeback.md

  when a project has

  memory/

  , component

  .agent/

  folders, or the user asks for cross-session memory.
• If the user asks about recent, concurrent, accepted, or current work, verify with current sources through web search, OpenReview, proceedings pages, arXiv, DBLP, Semantic Scholar, ACL Anthology, PMLR, CVF, or user-provided papers.
• If web access is unavailable, state that the output is a provisional reading plan and mark unverified papers or gaps explicitly.

Core Principles

• Optimize for project decisions, not an exhaustive bibliography.
• Separate canonical background, closest competitors, adjacent tools, and recent/concurrent threats.
• Treat unknown closest work as a major novelty risk.
• Rank papers by decision value: what changes the project if this paper is strong?
• Convert literature findings into baselines, ablations, claims to avoid, and writing positions.
• Preserve search provenance: where searched, when, which queries, and what was excluded.
• Do not overclaim novelty from absence of evidence.
• End with a next action that changes the project trajectory.

Step 1 - Define the Sprint Question

• topic, project idea, paper claim, or draft section
• target area and venues, if known
• intended contribution type
• known seed papers, baselines, datasets, or methods
• what decision the sprint must support
• time budget: quick scan, focused half-day, full sprint, novelty check, baseline check, or positioning check
• project memory IDs such as

  CLM-###

  ,

  RSK-###

  , or

  ACT-###

  , if present

For [topic/claim], determine whether [proposed contribution] is novel and important relative to [closest families], and identify [papers/baselines/gaps] that change the next project decision.

Step 2 - Build a Search Protocol

references/search-protocol.md

• seed concepts and synonyms
• method names and older terminology
• venue filters and likely communities
• canonical-source search paths
• recent/concurrent search paths
• backward and forward citation plan
• OpenReview or proceedings search plan when venue style matters
• stopping criteria

Step 3 - Collect and Classify Candidate Papers

references/paper-taxonomy.md

• foundational or canonical
• closest prior work
• direct competitor
• baseline method
• benchmark, dataset, or metric source
• adjacent method family
• theory or analysis source
• empirical survey or taxonomy
• recent or concurrent threat
• negative result or limitation evidence
• writing or positioning exemplar

Paper:
Role:
Core idea:
What it proves or demonstrates:
Relation to our project:
Decision impact:
Read priority:
Verification/source:

Step 4 - Prioritize Reading

references/reading-priority.md

• read-now

  : can change novelty, baseline selection, method design, or project viability

• skim

  : useful for context or framing, unlikely to change the core decision

• defer

  : relevant but not needed for this sprint's decision

• ignore-for-now

  : out of scope, weakly related, or superseded for current purpose

read-now

Step 5 - Synthesize the Literature Map

• method-family map
• chronology of key ideas
• closest-work comparison table
• baseline implications
• dataset, metric, or protocol implications
• theory or assumption implications
• open gaps and saturated claims
• terminology map for search and writing

• novelty-risk

  : close work may already cover the idea

• baseline-risk

  : a missing baseline would weaken experiments

• positioning-risk

  : the project is framed in the wrong community or contribution type

• evidence-risk

  : available experiments do not address the field's standard concern

• scope-risk

  : the literature is too broad for the current project shape

Step 6 - Convert Findings into Project Decisions

• should the idea be pursued, revised, parked, or killed?
• what is the closest prior work to beat or distinguish from?
• what claim is still defensible?
• what claim should be avoided?
• what baseline must be implemented or cited?
• what experiment, theorem, diagnostic, or analysis becomes mandatory?
• what writing frame is likely reviewer-friendly?
• what next skill should be used?

research-idea-validator

algorithm-design-planner

Step 7 - Write the Sprint Report

references/synthesis-template.md

docs/literature/literature_sprint_YYYY-MM-DD_<short-name>.md

• sprint question
• search log and limitations
• ranked paper map
• closest-work risks
• method taxonomy
• baseline and evaluation implications
• project decision implications
• next reading or experiment actions
• memory update section

Step 8 - Write Back to Project Memory

references/memory-writeback.md

• memory/decision-log.md

  : literature-driven project or positioning decisions

• memory/risk-board.md

  : closest-work, baseline, evidence, and positioning risks

• memory/action-board.md

  : read-now papers, baseline checks, implementation tasks, or writing tasks

• memory/claim-board.md

  : claims to keep, revise, narrow, park, or cut

• memory/evidence-board.md

  : planned baseline, dataset, metric, theorem, or diagnostic evidence

• paper/.agent/

  when related-work or positioning notes affect a draft

• verified

  for facts checked against primary sources

• user-stated

  for papers or constraints supplied by the user

• inferred

  for risk judgments and positioning implications

• unverified

  for search leads not yet checked

Final Sanity Check

• search scope and limitations are explicit
• papers are ranked by decision impact, not merely listed
• closest-work risk is named even if unresolved
• recent/concurrent search status is stated
• baseline implications are concrete
• project claims are adjusted when needed
• next skill or next action is unambiguous
• memory writeback is performed when the project has memory