aaai-submission — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited aaai-submission (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Use this for an AAAI main technical track submission audit. Reopen the current conference page, Author Kit, CFP, submission instructions, review process page, supplementary-material page, and author policies before giving deadline-ready advice.
Track-specific rules can differ.
metadata before the abstract and paper deadlines.
US Letter PDF, with a main PDF of at most 9 pages whose pages 8-9 are references only — 7 pages of non-reference content. Any ethical statement counts inside those 7 pages.
code/data, and metadata. Omit acknowledgments in the review version.
AAAI-27 moved it out of the document, and a checklist buried after the references is now a missing checklist.
combined main-technical-track submissions (the AI Alignment and AI for Social Impact special tracks are excluded from that count), allows no author additions after submission, and permits modifications only until the full-paper deadline.
under review at another archival venue, and AAAI-27 names the patterns it treats as violations: cosmetic re-submission, "thin slicing" one contribution across papers, splitting results by method/dataset/domain, and obfuscated or "alternative universe" duplicates. Papers sharing authors may be assessed together for novelty.
LLM-generated manuscript text, AI authorship, and AI-generated citations are policy risks.
AAAI's large reviewer pool and high volume mean a desk-level or Phase-1 cut is the most likely way to lose, so screen for the mechanical failures first.
| Check | Fast-fail trigger | Where it bites |
|---|---|---|
| Page limit | technical content over the cap | desk return |
| Anonymity | author name in PDF, code, or metadata | policy reject |
| Checklist | missing or contradictory | Phase-1 distrust |
| Dual submission | concurrent archival venue | ethics reject |
Build a single review-version packet before the abstract deadline, then refresh it before final submission. The packet is not a substitute for the official instructions; it is a way to make the submission audit reproducible and to keep late fixes from creating new policy problems.
| Packet item | What to capture | Failure it prevents |
|---|---|---|
| Policy snapshot | current conference page, Author Kit, submission instructions, supplementary-material rules, and AI-use policy checked dates | applying a closed cycle's facts to the open one |
| PDF proof | page count, paper size, style file, checklist placement, font/figure warnings, and metadata scrub result | desk return for formatting or anonymity |
| Author ledger | OpenReview profiles, conflicts, subject areas, author-limit count, and author-list freeze status | profile conflict, over-limit author, late author change |
| Supplement ledger | every appendix, multimedia, code/data ZIP, README, license note, and anonymous link policy decision | mutable web pointer or identity leak |
| Ethics ledger | dual-submission status, human-subjects/data constraints, AI-use disclosure, plagiarism/citation check | policy reject after submission |
For each item, record the evidence path and the person who can fix it. If a rule is uncertain, mark it as "requires current chair/author-kit confirmation" instead of guessing.
Run anonymity checks in a fixed order so fixes do not reintroduce leaks:
self-citations.
required.
executable logs.
keywords.
The output should separate "must fix before submission" from "acceptable but document why." Do not rely on a single manual skim; use PDF metadata tools, archive listing, and text search when available.
A robotics-learning team has 7 strong technical pages but left an acknowledgments line and a GitHub URL with their lab name in the supplement. The audit flags anonymity as the highest summary-reject risk: the fix order is strip the acknowledgment, replace the link with an anonymous archive, scrub ZIP metadata, then re-export and re-run the anonymity sweep before the paper deadline.
[AAAI readiness] Ready / Needs fixes / Not ready
[Track] Main / AI for Social Impact / AI Alignment / other
[Blocking checks] <page/anonymity/checklist/supplement/author-limit/dual-submission/AI-policy>
[Highest summary-reject risk] <one issue>
[Fix order] <ordered fixes before submission>~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.