Menu
Run fail-closed deterministic validation for mathematical and LaTeX artifacts
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/a5c-ai/babysitter --skill mathematical-artifact-validationThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Showing SKILL.md
| name | mathematical-artifact-validation |
| description | Run fail-closed deterministic validation for mathematical and LaTeX artifacts |
| allowed-tools | ["Bash","Read","Write","Glob","Grep"] |
| metadata | {"specialization":"mathematics","domain":"science","category":"mathematical-validation","phase":6} |
| graph | {"domains":["domain:mathematics"],"specializations":["specialization:computational-mathematics"],"skillAreas":["skill-area:mathematical-reasoning","skill-area:technical-writing"],"workflows":["workflow:research-validation","workflow:quality-convergence"],"roles":["role:research-scientist","role:computational-scientist"]} |
Produce reproducible source, structure, and reference evidence. Static validation is not mathematical proof verification.
Source paths, LaTeX artifact, optional BibTeX files, required sections, registry, tool policy, and output manifest paths.
validate_sources.py: strict UTF-8, nonempty files, no U+FFFD, SHA-256 manifest.validate_math_artifact.py: braces/environments, required sections, labels, references, theorem target numbering/type, citations.validate_registry.py: durable evidence and coverage invariants.validate_grade.py: rubric arithmetic/current hashes.Every task declares expectedExitCode: 0. Required-tool unavailability follows policy and is never represented as pass. An agent may classify diagnostics but may not override a failed shell result.
Use structured file paths, a run-owned workspace, Python standard library, and no Node. Never use lossy decoding. Never overwrite source or prior round artifacts.
Machine-readable gate manifest with pass|fail|unavailable, command/tool version where applicable, exit code, findings, evidence paths, and an explicit claim boundary.