Aaai26 Ai Review is a skill that generates structured, AI-assisted peer reviews of academic papers, modeled on the publicly described logic of the AAAI-26 AI Review Pilot. It operates as a multi-stage pipeline that first decomposes a paper into specialized analysis dimensions — contributions, methodology, experiments, and presentation — then synthesizes findings across those dimensions into a coherent draft. After synthesis, the skill runs a self-critique pass to flag unsupported claims in the draft, followed by a final quality check targeting anonymity leaks and citation hallucinations. This design mirrors the AAAI-26 philosophy of treating AI as an assistant to reviewers rather than a decision-maker: the output does not include an accept or reject verdict by default. The structured output format includes a Synopsis, Summary of Review, Strengths, a Weaknesses section with Major and Minor labels, and a References block. The skill accepts PDF input and is not restricted to AAAI submissions; it is designed to work for papers intended for NeurIPS, ICML, ACL, CVPR, and other academic venues. There are no environment variables required to configure it. It is published by agensi and is classified as a skill rather than an MCP server, meaning it runs as an agent capability rather than exposing individual callable tools.

Use cases

Generate a structured draft peer review for a machine learning paper submitted to NeurIPS or ICML

Identify unsupported claims in a draft review before a human reviewer finalizes it

Check a paper review draft for accidental anonymity leaks or hallucinated citations

Decompose a paper's contributions, methodology, and experiments into separate structured assessments

Produce a review following the AAAI-26 structured format (Synopsis, Strengths, Weaknesses with Major/Minor labels)

Assist program committee members by generating a review baseline from a PDF submission

When to use it

When a human reviewer wants a structured first draft to react to rather than write from scratch

When processing PDF submissions for academic venues that expect AAAI-style structured reviews

When the workflow explicitly avoids automated accept/reject decisions and treats AI as an assistant

When citation hallucination and anonymity preservation checks are required before a review is submitted

When not to use it

When an automated accept/reject verdict is required — the skill does not produce one by default

When input is not in PDF format, as no other input formats are specified in the facts

When a fully deterministic, auditable review pipeline with version-pinned packages is required — no package version is available

When the deployment environment requires an MCP server with callable tools rather than a skill-based agent capability