SoK: Agentic Skills
A systematization of reusable skills beyond tool use in LLM agents.
PhD Researcher · Casual Academic · UTS CybeR Lab
Delong Li
I study machine unlearning, agentic AI security, privacy-aware inference, and auditable model updates, with a focus on how adaptive AI systems can change without compromising trust.
Research updates
Recent publications, open-source releases, and research milestones.
Academic Paper Skills
A four-skill Codex bundle for policy, writing, figures and tables, and review.
AuditableLLM published in Electronics
A hash-chain-backed framework for compliance-aware LLM update auditing.
Latest 3 updates
Publications · Open source · Research
Research focus
I examine how AI systems can be updated, deployed, and evaluated under real-world security, privacy, and governance constraints.
Research perspective
AI systems should be able to change without becoming harder to trust.
My work brings together machine unlearning, privacy-aware inference, agentic security, and auditable model updates. I develop methods and evaluations that make system changes easier to inspect, explain, and govern.
Verifiable updates
Develop model adaptation and unlearning methods whose effects can be traced and evaluated.
Privacy in practice
Design federated and edge-cloud inference methods for practical privacy and deployment constraints.
Evidence-based trust
Assess system behavior through security analysis, empirical evaluation, and audit evidence.
Publications
Selected work on machine unlearning, agentic AI security, privacy-aware inference, and auditable model updates.
Featured publications
2 featured
SoK: Agentic Skills -- Beyond Tool Use in LLM Agents
A systematization of agentic skills across their lifecycle, with taxonomies for packaging, representation, execution, evaluation, and governance risk.
Agentic systems
Skills
Governance
A Survey of LoRA-based Machine Unlearning for LLMs: Methods, Taxonomy, and Evaluation
A survey of adapter-centric machine unlearning for LLMs, organizing methods through a unified adapter-space taxonomy and evaluation view.
Machine unlearning
LoRA
Survey
Additional publications
By year
Projects and research tools
Open-source tools and reusable resources for research, academic writing, and evaluation.
In development
Featured project
Codex skill bundle
Academic Paper Skills
A coordinated bundle of four Codex skills for policy-aware writing, manuscript rule auditing, publication-ready figures and tables, and pre-submission review.
- Structure
- Four Codex skills
- Focus
- Academic workflows
- Access
- Open source
Project index
Actively maintained
Paper Visualization
A reusable skill for turning paper PDFs, preprint pages, or source workspaces into readable static research visualizations.
Research and collaboration
Get in touch
I welcome conversations and collaborations on machine unlearning, agentic AI security, privacy-aware inference, and auditable AI.
Machine unlearning
Agentic security
Auditable AI