dechenzhang/valg-ml-theory-agent/proof-assembly
Claude Code and agent skill by dechenzhang/valg-ml-theory-agent
Trust tier 2
- community
Source check pending
Checking that the SKILL.md file is still in the repository.
Assemble accepted proof-step artifacts into a self-contained, paper-ready LaTeX bundle without introducing unreviewed lemmas.
Install
npx vskill@latest install dechenzhang/valg-ml-theory-agent/proof-assemblySKILL.md
Proof Assembly
Use this internal skill after all required proof steps have accepted reviews.
Contract
Required inputs:
perspective_M/idea_N/setting.md- accepted
perspective_M/idea_N/proof_sketch.md - accepted
perspective_M/idea_N/proof_sketch_review.md - accepted
proof_steps/<Step ID>/proof.mdandproof_steps/<Step ID>/review.mdfor every required sketch step
Optional input:
- final
proof_review.mdwhen repairingPROOF_ASSEMBLY_FLAW
Outputs:
perspective_M/idea_N/latex_template/1_intro.texperspective_M/idea_N/latex_template/2_preliminary.texperspective_M/idea_N/latex_template/3_main.texperspective_M/idea_N/latex_template/4_proof_sketch.texperspective_M/idea_N/latex_template/5_appendix.texperspective_M/idea_N/latex_template/main.texperspective_M/idea_N/latex_template/arxiv.styperspective_M/idea_N/latex_template/ims.bstperspective_M/idea_N/latex_template/reference.bibperspective_M/idea_N/assembly_report.md
Do not modify sketch artifacts, step artifacts, trackers, worker logs, or accepted results.
Responsibilities
- Assemble only accepted step proofs into theorem-level form.
- Preserve accepted step statements, assumptions, and dependencies.
- Prove the final theorem by composing accepted step claims and direct assembly-only implications.
- Write
latex_template/as the canonical public proof source. In the LaTeX bundle, "paper-ready" means polished theorem-style exposition: compact exposition for the main theorem and proof sketch, but complete journal-style proof detail forlatex_template/5_appendix.tex, not a condensed appendix summary. - Write
assembly_report.mdas the controller-private assembly status artifact. Do not put workflow status fields such as## Status,COMPLETE, orPARTIAL_BLOCKEDinto public TeX. - Stop honestly when assembly needs an unreviewed lemma, changed step statement, changed assumption, or new proof obligation.
Workflow
Step 1: Load Accepted Inputs
- Read the exact formalized setting and goal from
setting.md, including the stableassump:<slug>ids assigned to technical assumptions. - Read the accepted
proof_sketch.mdandproof_sketch_review.md. - Read every accepted
proof_steps/<Step ID>/proof.mdandproof_steps/<Step ID>/review.mdrequired by the sketch. - If repairing from final
proof_review.md, repair only assembly-level defects, notation reconciliation, presentation, or LaTeX-bundle synchronization. Do not repair step or sketch defects inside assembly. - If repairing
PROOF_ASSEMBLY_FLAWthat reports compressed, missing, or unsynchronized appendix proof detail, regenerate each affected appendix subsection from the corresponding accepted proof-step source body. Do not use the compressed appendix as the proof source and do not try to polish a representative summary into acceptance. Severe shortening is repairable only after direct comparison with the accepted source sections shows that the removed text is non-proof cleanup, exact duplicate setup, or valid repetition under the repetition rule.
Step 2: Verify Coverage And Dependency Closure
- Confirm every required sketch step has an accepted proof artifact and an accepted review artifact.
- Confirm every step dependency used by an accepted step is also accepted and available.
- Treat the accepted
proof_steps/<Step ID>/proof.mdfiles as the source of truth for appendix derivations. For every used accepted step, perform an internal proof-obligation pass over that step's## Local Derivation, all invoked## Cited Result Applications, and## Target-Step Assembly. - Preserve every mathematical proof-obligation item from that pass in the matching
latex_template/5_appendix.texsubsection unless it is exact duplicate setup, non-mathematical repetition, or a genuinely repeated derivation represented by one complete instance plus exact substitutions for each abbreviated instance. - Removing workflow labels, proof-flow headings, provenance phrases, audit scaffolding, sketch-row fields, or dependency inventories never authorizes deleting the mathematical content beneath them. Rewrite the mathematical content as theorem-style statements, proof text inside proof environments, displayed derivations, cited-result instantiations, or assumption-discharge text; omit the audit packaging itself from public TeX.
- If a sketch step is unnecessary for the final theorem, record the reason in the assembly text; do not silently drop it.
- If a required step, local unit, cited-result application, boundary case, assumption discharge, target-step assembly, source unit, or dependency is missing, stale, rejected, unresolved, or cannot be translated into the appendix, keep the bundle honest, do not present the target as proved, and record the blocker in
assembly_report.md.
Step 3: Set Attempted Claim And Assembly Report
- Set
assembly_report.md## StatustoCOMPLETEonly when the accepted step claims and direct assembly-only implications prove the formalized goal in a reviewable way. - In exact-goal mode,
COMPLETErequires the attempted claim to match the exact target claim fromsetting.md. - In target-spec mode,
COMPLETErequires the attempted claim to be a concrete valid instantiation of the target quantity, claim type, active scope, success criterion, and required dependence. - If required dependence is explicit in the goal, do not leave placeholder dependence such as "up to constants" or "up to polynomial factors".
- If a needed derived invariant is not proved by accepted steps or direct assembly-only implications, set
assembly_report.md## StatustoPARTIAL_BLOCKEDand name the missing Assumption Provenance Bridge. Do not repair by adding the derived invariant to the main theorem assumptions unless the formalized goal is explicitly conditional. - If the assembly proves only a weaker statement, use that exact weaker attempted claim and set
assembly_report.md## StatustoPARTIAL_BLOCKED. - If blocked, keep valid assembled content and name the exact failed target theorem, accepted step block, missing coverage item, notation reconciliation, dependency closure, or final implication in
assembly_report.md## Blockers.
Step 4: Write LaTeX Bundle
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/proof-assembly/SKILL.md
Popularity
- Repository stars
- 70
- Repository forks
- 3
- 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
[](https://verified-skill.com/skills/dechenzhang/valg-ml-theory-agent/proof-assembly)