sakanaai/ai-scientist-v3/validate-paper

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Validate grounded paper artifacts and compile LaTeX PDF with strict fail-fast checks.

Install

npx vskill@latest install sakanaai/ai-scientist-v3/validate-paper

SKILL.md

What the skill is about

Run final quality gate for the generated paper.

When to use the skill

  • After paper.tex, references.bib, and claims_grounding.json are written.
  • Before declaring final paper success.

What to do

  1. Validate sections and citation keys:
bash
python scripts/validate_tex_refs.py \
--tex paper.tex \
--bib references.bib

Require a substantive embedded Supplementary Information section after the bibliography. It must use \\appendix and \\section{Supplementary Information}, contain grounded visible content plus at least one subsection, table, or figure, and contain no template placeholder text.

  1. Validate claim mappings, bibliography/provenance consistency, and every @evidence[[...]] link in the scientific-evidence/ bundle:
bash
python ../write-paper/scripts/validate_claims.py \
--paper-data paper_data.json \
--references references_provenance.json \
--bib references.bib \
--claims claims_grounding.json \
--paper-dir .

This delegates to the scientific-evidence skill verifier, which classifies every @evidence[[...]] link as resolved, anchored, or dangling. Any dangling link (or an @evidence[[...]] token with no scientific-evidence/ bundle present) is a hard error and must be sent back to the paper-writing agent for repair.

  1. Validate figure manifest and figure readability budget when figure_manifest.json is present or paper_strategy.json requires figures. The canonical schema is templates/figure_manifest.schema.json.
bash
python .claude/skills/validate-paper/scripts/validate_figures.py \
--paper-dir .
  1. Compile:
bash
bash scripts/compile_latex.sh paper.tex

The compile script also validates that Supplementary Information is visibly rendered on a page after References with substantive content. Source-only structure checks are not sufficient for conditional or literal LaTeX constructs. It records and validates every LaTeX input, restricts BibTeX to the canonical references.bib, and removes inherited Kpathsea path overrides before compiling.

  1. Require paper.pdf to exist.
  2. When paper_data.json.paper_profile == "standalone", also fail on visible paper language that leaks pipeline terms such as eligible, correct and eligible, correctness rate, worker_01, worker_02, combined_score, or similar internal wording.
  3. When the selected format or venue guidance requires real author metadata, also fail on anonymous or generic placeholder author blocks and emit soft warnings when title/abstract/introduction/conclusion lack venue-relevant framing or when the discussion drifts into unrelated application narratives.
  4. Fail hard on any missing grounding link, missing section, missing/empty/misordered Supplementary Information, unresolved citation, bibliography/provenance mismatch, figure-manifest mismatch, density-budget violation, page-limit violation for formats with enforced budgets, standalone-language violation, author-block violation required by format/venue guidance, dangling @evidence[[...]] link, or compile error.
  5. Treat the parent orchestrator's validation report as authoritative during repair rounds. Last-body-page fill is rendered geometric vertical occupancy computed from text and graphics bounds, not character count or word density. Do not claim that page validation passed using a character-count proxy. If the canonical validator is unavailable in the agent workspace, make the targeted repair and return control so the parent can compile and revalidate it.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
437
Repository forks
67
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/sakanaai/ai-scientist-v3/validate-paper/badge)](https://verified-skill.com/skills/sakanaai/ai-scientist-v3/validate-paper)

Something wrong with this skill or its page? Tell us.