Dechen Zhang
Community@DechenZhang · Hong Kong
PhD student in HKU
Agent Skills by Dechen Zhang
Showing 17 vetted skills indexed across 1 GitHub repositories.
DechenZhangproof-review-aggregate
Aggregates four specialized proof reviews into a final controller-facing proof_review.md with score and routing.
DechenZhangidea-formalizer
Formalizes approved research ideas into theorem-ready setting.md artifacts with explicit assumptions and goals.
DechenZhangproof-step
Proves one accepted proof-sketch step with local lemmas, checked citations, and target-step assembly.
DechenZhangproof-review-citation
Reviews final proofs for citation fidelity, object mapping, and assumption discharge.
DechenZhangproof-review-structural
Reviews assembled mathematical proofs for goal alignment, dependency closure, and LaTeX structural consistency.
DechenZhangglobal-proof
Drafts a theorem-level diagnostic proof from an accepted proof sketch before step-level proof work.
DechenZhangproof-sketch
Generates branch-local proof sketches with lemma-sized steps, dependencies, and assumptions for theorem proving.
DechenZhangproof-sketch-review
Reviews branch-local proof sketches as an early viability gate before step-level proof work.
DechenZhangperspective-select
Convert literature gaps into normalized ML-theory perspectives and write Perspective_Selection.md.
DechenZhangml-theory-research-agent
Orchestrates multi-stage ML theory research workflows from literature survey to verified theorem proofs.
DechenZhangsubagent-idea-generator
Generates one branch-local theorem idea under an approved perspective for ML theory research.
DechenZhangglobal-proof-review
Reviews global proof diagnostics and routes failures to the smallest valid repair stage.
DechenZhangproof-review-rigor
Audits assembled mathematical proofs for derivational rigor, quantifiers, constants, and boundary cases.
DechenZhangproof-review-adversarial
Stress-tests theorem-level and step-level proof claims for counterexamples and hidden gaps.
DechenZhangproof-step-review
Reviews one proof-step artifact with per-lemma audits, hidden subclaim scans, and rigor checks.
DechenZhangproof-assembly
Assemble accepted proof-step artifacts into a paper-ready LaTeX theorem bundle.
DechenZhangresearch-lit
Maps ML theory literature and writes a structured LITERATURE_SURVEY.md with gaps.
Frequently Asked Questions About Dechen Zhang
FAQPage SchemaWhat tasks can I automate using Dechen Zhang's skills?▼
You can run the full ML theory research pipeline: survey literature, select perspectives, generate and formalize ideas, build proof sketches, prove individual steps with local lemmas, run split-mode diagnostic reviews (structural, rigor, citation, adversarial), and assemble accepted steps into a self-contained LaTeX proof bundle.
Who are these skills designed for?▼
They target ML theory researchers and PhD students, such as the author's HKU research context, who use Codex to formalize ideas, draft theorem-level proofs, and enforce rigorous multi-stage review gates before accepting proof artifacts.
How does the proof verification pipeline work in practice?▼
A proof sketch is gated by proof-sketch-review, then global-proof writes a diagnostic draft reviewed by global-proof-review. Each step is proved and audited via proof-step-review, and four specialized final reviewers feed proof-review-aggregate, which emits the final proof_review.md with score and next action.
What are the prerequisites and dependencies for these skills?▼
Skills operate on branch-local artifacts (perspective_M/idea_N directories) within the ml-theory-research-agent orchestration, requiring a Codex environment, an approved perspective from perspective-select, and prior-stage outputs such as LITERATURE_SURVEY.md or an accepted proof sketch.
Are Dechen Zhang's skills open source and free to use?▼
The skills are published as a public registry manifest under the DechenZhang account with no stated licensing fees or commercial terms in the manifest. Users should check the repository for an explicit license before redistribution or derivative use.