Recent AI news · Original report: 2026-09-15 (KST)

GitHub AI skills roundup: explainer videos, screenwriting, NVIDIA SkillSpector (Sep 15, 2026)

This article was compiled with AI. This English translation preserves the original report's collection date, findings, and limitations. Please consult the linked sources for current information. Originally collected: 2026-09-15 13:42 (KST, UTC+09:00).

Translated into English: 2026-09-28 (KST). The translation date does not indicate a new source check.

At a glance

This article summarizes three AI skills created between September 6 and 9, KST, that had passed 1,000 stars, along with NVIDIA SkillSpector, which says it scans skills before installation. The descriptions are based on their READMEs. We did not run them ourselves.

Repository overview

These figures were retrieved through the GitHub API on September 15 at 1:42 p.m. KST. Latest commit dates are also in Korean Standard Time.

anything2explainer: a topic becomes an explainer video

The README describes a Claude Code and Codex skill that takes a topic and creates an explainer video with motion graphics, voice narration, subtitles, and a chapter progress bar. It says visuals are drawn with Remotion code, without stock footage or video-generation models, and production takes around one to three hours depending on the length.

• License: The API labels the license “Other,” but the LICENSE file specifies PolyForm Noncommercial 1.0.0. It says commercial use requires the author’s prior permission.
• The README says it has not been tested on Windows.

screenwriting-skills: 26 screenwriting skills

The repository describes 26 skills drawn from 47 books on writing craft and 23 published scripts, plays, and related volumes. Counting the skill folders through the API also gives 26. The skill text is in Chinese, but the README says it answers in Korean when asked in Korean, and there is a Korean README. The author removed the English version, explaining that separate translations caused the content to diverge between languages. The author also acknowledges the resulting limitation: people who cannot read Chinese cannot directly review the original text.

dream-loop: create a target image, then work toward it

The README describes the following process:

• Use image generation to create a target image first.
• Build a game, app, or scene using that target as a reference.
• Have a separate AI critic compare the actual screen with the target and give feedback.
• Repeat until the critic is satisfied.

It says image generation and image-input capabilities are required, and Blender is used for 3D modeling. At the time of the article, the README said it had been tested only with GPT-6 Astra in Codex.

SkillSpector: scan a skill before installation

The README describes a scanner for finding prompt injection, data leakage, supply-chain risks, and other issues before installing Claude Code, Codex, or MCP skills. It combines static analysis with optional LLM analysis to produce a risk score from 0 to 100. The repository was created in March, but it was included here because it also had commits this week.

• Data transmission: With LLM analysis enabled, which is the default, file contents are sent to the configured model provider. The README says that even in static-analysis-only mode, dependency names and versions are sent to OSV.dev for vulnerability lookups.
• Stated limitations: It says it is not a sandbox and can miss non-English patterns, text in images, and encrypted code.

What we could not verify

• Actual operation and security: We did not install, run, or conduct a code security review of any of the four repositories.
• The rate of star growth: The table shows cumulative stars at the time of the query. We did not measure growth.
• The number of SkillSpector patterns: The README body says 71, but counting the items in its table gives 72.
• Statistics in the SkillSpector README: It cites a paper by Liu et al. (2026) for the claim that 26.1% of 31,132 analyzed skills contained vulnerabilities. We did not read the paper itself.
• Original sources and compatibility: Beyond the README descriptions, we did not verify whether screenwriting-skills accurately reflects its stated sources or whether dream-loop works with other models.

Sources

• GitHub, Vincentwei1021/anything2explainer https://github.com/Vincentwei1021/anything2explainer
• anything2explainer LICENSE https://github.com/Vincentwei1021/anything2explainer/blob/main/LICENSE
• GitHub, jtydhr88/screenwriting-skills https://github.com/jtydhr88/screenwriting-skills
• screenwriting-skills Korean README https://github.com/jtydhr88/screenwriting-skills/blob/main/README_KO.md
• GitHub, achimala/dream-loop https://github.com/achimala/dream-loop
• GitHub, NVIDIA/SkillSpector https://github.com/NVIDIA/SkillSpector