Agent Skills
› NeverSight/learn-skills.dev
› paper-evidence-board
paper-evidence-board
GitHub维护面向论文的论据看板，对齐主张、实验数据、图表与章节，评估审稿风险并规划提交前行动。适用于写作中暴露证据缺失、需更新叙事、模拟审稿或检查主张支持度等场景。
Trigger Scenarios
写作发现论文主张缺乏证据
新实验结果要求更新论文叙述、图表或章节
审稿模拟产生实验或写作任务
提交前需要检查主张与证据的对齐情况
需要将图表映射到论文主张
用户希望生成实时的论文论据看板
Install
npx skills add NeverSight/learn-skills.dev --skill paper-evidence-board -g -y
SKILL.md
Frontmatter
{
"name": "paper-evidence-board",
"description": "Maintain a paper-facing evidence board that aligns claims, experiments, figures, tables, sections, reviewer risks, and next actions during ML\/AI paper writing. Use this skill whenever writing exposes missing experiments, new results require paper changes, reviewer simulation reveals evidence gaps, claims need support checks, figures\/tables need mapping to claims, or the user wants a live paper evidence board before submission.",
"allowed-tools": "Read, Write, Edit, Bash, Glob, WebSearch, WebFetch",
"argument-hint": "[paper-dir-or-project-root] [--venue <venue>] [--mode build|audit|update|review]"
}
Paper Evidence Board
Maintain the paper-facing view of the research project: which claims the paper makes, what evidence supports them, where they appear, what reviewers may attack, and what actions are needed before submission.
Use this skill when:
- writing reveals that a paper claim lacks evidence
- new experiment results require updating paper narrative, figures, or sections
- review simulation creates experiment or writing actions
- a paper draft needs claim-evidence alignment before submission
- figures and tables need to be mapped to paper claims
- figure/table style needs to be tracked as a writing decision, including palette, markers, symbols, typography, and venue-facing visual conventions
- the user wants a live evidence board for a paper
- the project needs to decide whether a gap requires more experiments, rewriting, narrowing a claim, or cutting material
This skill is more specific than research-project-memory: it uses project memory as the source of cross-component truth, but creates a paper-facing board for writing and review decisions.
Pair this skill with:
research-project-memoryfor project-level claim/evidence/risk/action IDs and writebackconference-writing-adapterwhen the board shows structural or paragraph-level writing changesexperiment-design-plannerwhen the board exposes missing evidence requiring new experimentsresult-diagnosiswhen new evidence weakens or complicates a claimpaper-reviewer-simulatorwhen the board should be stress-tested from a reviewer perspectivefigure-results-reviewwhen figures/tables need claim-support, caption, statistical, or visual-style reviewcitation-coverage-auditwhen novelty or related-work gaps appear
Skill Directory Layout
<installed-skill-dir>/
├── SKILL.md
└── references/
├── board-schema.md
├── evidence-gap-triage.md
├── paper-section-map.md
├── report-template.md
├── reviewer-risk-integration.md
└── update-protocol.md
Progressive Loading
- Always read
references/board-schema.md,references/evidence-gap-triage.md, andreferences/update-protocol.md. - Read
references/paper-section-map.mdwhen building a section-by-section board from a draft. - Read
references/reviewer-risk-integration.mdwhen using reviewer simulation, real reviews, or venue-specific risks. - Use
references/report-template.mdfor substantial board reports. - Verify current venue rules or reviewer forms when venue-specific evidence expectations matter.
Core Principles
- Every major paper claim should have a paper location, evidence status, and reviewer risk.
- Evidence should point to source artifacts, not only prose memory.
- A figure/table should have a job: support a claim, answer a reviewer question, or delimit scope.
- A paper-facing figure/table should also obey the paper's visual style policy, not just display correct numbers.
- Missing evidence is not a writing problem by default; it may require experiment design, result diagnosis, claim narrowing, or citation work.
- Do not hide negative or weak results. Mark how they change the claim.
- Keep the board actionable: each open gap should route to a next skill or action.
Step 1 - Locate Paper and Project Memory
Find:
- paper root:
paper/, current directory, or user-provided path - project root, if different
memory/claim-board.mdmemory/evidence-board.mdmemory/risk-board.mdmemory/action-board.mdpaper/.agent/paper-status.md- paper sources such as
main.tex,paper.tex,sections/*.tex, figures, tables, and appendix
If project memory is missing, still build a paper board from the draft, but recommend initializing research-project-memory.
Step 2 - Extract Paper Claims
Read the draft or provided notes and extract:
- title and abstract claims
- introduction contribution bullets
- method claims
- theory claims
- experiment claims
- related-work novelty boundaries
- limitation and scope claims
Assign or reuse CLM-### IDs. For each claim, record:
- exact paper location
- claim type
- strength
- current wording risk
- evidence expected
- evidence available
Do not treat aspirational contribution language as proven evidence.
Step 3 - Map Evidence to Claims
Read references/board-schema.md.
For every evidence item:
- identify source path, run ID, table, figure, theorem, citation, or review text
- link to claim IDs
- mark status: planned, running, observed, reported, stale, contradicted, or cut
- mark certainty: observed, user-stated, inferred, stale, or needs-verification
- state limitations
If evidence is only in a daily log or chat note, mark it accordingly and create an action to verify source artifacts.
Step 4 - Build Section and Figure/Table Map
Read references/paper-section-map.md.
Create:
- section map: section -> claims -> evidence -> risks -> actions
- figure/table map: figure/table -> evidence -> claim -> required caption message -> stale status
- visual style map: palette, method-to-marker mapping, typography, symbols, figure sizing, and table conventions
- appendix map when relevant
Identify:
- claims repeated in multiple places with inconsistent strength
- figures without a claim
- claims without a figure/table/proof/citation when one is expected
- stale figures after new results
- inconsistent colors, symbols, method names, font sizes, or figure sizes across paper visuals
- result tables not discussed in prose
Step 5 - Triage Evidence Gaps
Read references/evidence-gap-triage.md.
For each gap, decide:
new-experiment: needsexperiment-design-plannerresult-diagnosis: existing result is ambiguous or contradictoryrewrite: evidence exists but prose overclaims or hides itnarrow-claim: evidence supports a smaller claimcitation-work: claim needs missing related work or attributioncut: claim is not worth supportingaccept-risk: limitation can be stated rather than fixed
Every high-risk gap should create an action.
Step 6 - Integrate Reviewer Risks
Read references/reviewer-risk-integration.md.
Use:
- simulated reviewer risks from
paper-reviewer-simulator - citation risks from
citation-coverage-audit - real review risks from
rebuttal-strategist - venue-specific expectations when known
Map each risk to:
- claim
- paper location
- evidence gap
- fix type: experiment, analysis, proof, citation, rewrite, figure/table, limitation, or rebuttal
- priority
Step 7 - Produce Board and Actions
Use references/report-template.md for full output.
If saving to a project and no path is given, use:
paper/.agent/paper-evidence-board.md
Output:
- paper snapshot
- claim-evidence matrix
- section map
- figure/table map
- visual style map
- evidence gaps
- reviewer-risk map
- action plan
- project-memory writeback
Step 8 - Write Back to Project Memory
If the project uses research-project-memory, update:
memory/claim-board.md: claim status, wording, paper locations, and weakened/cut claimsmemory/evidence-board.md: evidence source paths, figure/table mappings, stale status, and limitationsmemory/risk-board.md: evidence-gap, writing, novelty, baseline, and reviewer risksmemory/action-board.md: experiment, diagnosis, rewrite, citation, figure/table, and review actionspaper/.agent/paper-status.md: current paper section statepaper/.agent/visual-style.md: paper-facing figure/table style policy when visual conventions are materialpaper/.agent/paper-evidence-board.md: the paper-facing board
Do not duplicate long experiment reports. Link to them.
Final Sanity Check
Before finalizing:
- every major claim has a location and evidence status
- unsupported claims are marked as planned, narrowed, or cut
- every main figure/table has a claim-facing job
- paper-facing figures/tables share consistent colors, markers, symbols, typography, sizing, and notation
- stale evidence is marked
- reviewer risks link to actions
- next actions are routed to the right skill
- project memory is updated when present
Version History
- e0220ca Current 2026-07-05 21:35
