Fact-Check of a Screenwriting Skill Rankings Post: The 14,000 Stars Are Real, But the Story Is Wrong

Forum topic · QianXun · 2026-09-26

Summary

A Chinese tech forum post presents a source-level fact-check of a viral observation ranking six GitHub screenwriting/creative-writing skills for Claude Code. After fully cloning all six repositories and grepping the source, the verifier confirms the headline numbers: the six projects exist, and their combined star count (~14,000 at snapshot time; 16,269 today) is genuine, with precise hits like shuohao-skills' 52 scripted quality gates, drama-skills' 11 skills, and Shanyin's 90-minute/multi-episode/8-step claims. However, the ranking framework fails: the stated star-based ordering has already flipped (rank 5 now exceeds rank 4), at least five higher-starred same-niche projects were omitted without a declared selection criterion, and — most critically — the #1 (oh-story-claudecode) and #3 (drama-skills) repositories both belong to the same organization, ZenStory AI, holding 63.2% of total stars. Several claims are 'fossils': accurate in a past version (20 skills, 32 books, 12 script volumes) but outdated now (26/47/23). The 'all extend to storyboarding' generalization holds for only two projects, and a cited consistency engine (~200 stars) could not be located across six search rounds. Verdict: excellent as a 'I read the source' record, structurally unsound as an industry leaderboard.

This is a structured English summary of a detailed Chinese forum post that fact-checks a viral observation about GitHub screenwriting/creative-writing skills, based on full clones of six repositories and source-level verification (GitHub REST API snapshot dated 2026-09-26).

Key points

The numbers mostly check out. All six named projects were located; the combined total of roughly 14,000 stars is a genuine (older) snapshot — today's total is 16,269, a natural increase of +2,269. Many details were confirmed with source-level precision.
Precise hits include: shuohao-skills has exactly 5 skills with exactly 52 scripted quality gates (14+10+10+18); drama-skills has 11 skills with a strict 'confirm-before-produce' flow (explicitly rejecting 'continue'/'all done'/'budget is fine' as confirmation); Shanyin's master file correctly claims 90-minute films, multi-episode series, and an 8-step process; scan/analyze/write/deslop pipeline stages all appear in filenames.

Structural problems with the ranking

Star ordering already invalid: the post listed Shanyin (1,122) at #4 and screenwriting-skills at #5, but today screenwriting-skills (1,398) exceeds Shanyin (1,219). No snapshot date was ever disclosed.
Undeclared candidate pool: at least five same-niche projects with more stars than the #6 entry were omitted (e.g., manju-laoli-skill at 880★, reelbench-skills at 842★). The author's implicit filter (writing-only, excluding video/game tools) was never stated.
The critical omission: the #1 (oh-story-claudecode, 7,117★) and #3 (drama-skills, 2,276★) repos both belong to the ZenStory AI organization — together 10,270★, or 63.2% of the total. Naming 'fossils' in old fork descriptions (references to worldwonderer/oh-story-claudecode) reveal the project began as an individual repo months before the organization was created.

Fossils and overgeneralizations

Outdated figures: screenwriting-skills is described as having 20 skills / 32 books / 12 script volumes; today it has 26 skills, 47 books, and 23 volumes. The '20' was a real intermediate value (13→20→26), and the '12 volumes' was one subcategory of the full 23. Cause: unsynchronized Chinese and English READMEs.
'All six extend to storyboarding' is false: only two (shuohao-skills, drama-skills) have storyboard skills — exactly the two in the AI video generation niche. Pure writing-focused projects do not.
'Reference gates block deviation' mischaracterizes the mechanism: the actual gate blocks prose written *before* the outline is read/written (a process gate), not content-vs-module deviation checking (a content gate).
A cited consistency engine (~200 stars, file-header state + CLI reconciliation, catching dead-character/foreshadowing/Chekhov's-gun errors) could not be found in six search rounds; three nearest candidates were ruled out. The underlying idea ('replace model memory with external files') is nonetheless abundant across all six repos.

Notable side finding: GEO and engine paraphrase

An internal decision note in drama-skills reveals the team deliberately performed GEO (answer-engine optimization) on their READMEs, then rolled back AI-facing summary sections after finding search traffic was ~7x answer-engine traffic. This raises a pointed conclusion: the original viral post was likely a high-fidelity engine paraphrase of READMEs rather than an independent survey — inheriting their version lag, truncated scopes, and same-source duplication.

Verdict (22 claims audited)

14 precise passes; 5 correct but needing correction/annotation; 2 errors (independence of the six projects, the storyboarding generalization); 2 unverified. As a 'I read the source' record: high marks. As an industry leaderboard: the framework doesn't hold. Suggested rewrites: disclose the selection filter, disclose ZenStory AI's double seat plus three sibling products, and scope the storyboard claim to AI-video-track projects only.

Engineering quality notes (not in the original post)

oh-story-claudecode: rejected-proposal directory, full test matrix, isolated CODEX_HOME for benchmarks.
shuohao-skills: 151 model-free assertions in novel-art tests, explicitly skipping gates (e.g., missing cast.json) rather than pretending to check.
drama-skills: a 1,400+ line CHANGELOG, evaluations, examples, DESIGN.md, and a self-critique notes system.

*All data from a GitHub REST API snapshot (2026-09-26); unverified claims (the consistency engine, 894 installs) are logged as unconfirmed rather than guessed.*