# A Bilibili video claimed "tokens cut by 27x," and a DeepSeek plugin got 5300 stars in 29 days · GitHub open-source project

> Source: https://stringers.io/en/github/gh_T-Auto__dsh-ops.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> The author posted two short Bilibili videos, "saves 40%" and "27x optimization" (250k and 380k views), and the star peak lined up with the release dates. His dsh-TUI team cross-promoted it, and the desktop client bundles it directly.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

T-Auto/dsh-ops · T-Auto

# A Bilibili video claimed "tokens cut by 27x," and a DeepSeek plugin got 5300 stars in 29 days

**What it is** A DeepSeek Harness plugin that swaps PowerShell for bash on Windows and adds a set of file tools returning more compact results, so the AI makes fewer errors and burns fewer tokens.

**Why it took off** The author posted two short Bilibili videos of his own, "saves 40%" and "optimized 27x" (250k and 380k plays). The star peak lined up with the release day. His dsh-TUI team cross-promotes, and the desktop client bundles it directly.

**B** **How you could follow** Stop building generic token-saving plugins. Look in agent tools that just opened up plugins for a specific pain point nobody has claimed, port a tool already validated in another ecosystem, back it with a measurable number and a short Bilibili video, and get adjacent plugin authors to cross-promote.

Grade B: the playbook is clear and easy to copy, but the token-saving niche already has a dozen-plus plugins. Its growth came from Bilibili's existing audience, a team recommended as official, and bundling in the client, which latecomers can't get in the short term. Spreading in small circles

Data (updated daily) · Repo created 2026-10-08 · 5,515 stars · 7 days +580

GitHub · stars

Data 2026-10-10 → 2026-11-08 (30 days, 30 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

dsh-ops is a plugin written for DSH, DeepSeek's official agent tool, that does one small thing: it makes the AI type fewer wrong commands and burn fewer tokens on Windows. It invents nothing new. Its stars came half from hitting a real Windows-user pain point and half from its author being a Bilibili UP host himself. Below: why it took off, where the stars came from, whether it's still worth following now, and how to follow it.

### a 3-and-a-half-minute video

In Oct 2026 14, Bilibili UP主 "魔法师風雪" and a co-creator posted a 3-minute 31-second video titled "Did you know your DeepSeek can be optimized to cut token consumption by 27x? Stop wasting tokens!" The video description linked to a GitHub repo, T-Auto/dsh-ops, which had been created just 6 days earlier.

By early November the video had 384k plays. On the repo side, stars rose by 10-15 on 584 day and by 10-16 on 799, the biggest single-day gain since the repo was created. His first video, "A one-click plugin that saves 5% of DSH token spend!" posted 40 days before, also had 250k plays, and stars started climbing the day after it went up.

By Nov 6, day 29 since the repo was created: 5298 stars and 188 forks. The same-named npm package went from 10-08 to 11-05 with about 144k downloads (npm's official download stats).

### What it actually is

DSH is an agent tool that DeepSeek open-sourced in Aug 2026 13 (DeepSeek Harness), built around "everything is a plugin." Many people run it on Windows, and that's where the trouble starts: DSH uses PowerShell by default on Windows, but DeepSeek models tend to write bash-style commands. Those commands error out, the agent retries, the retries error out again, and the tokens burn away.

dsh-ops fixes this by swapping the tools the AI uses, without touching the context:

- It bundles a full bash on Windows (based on the open-source project niubash) to replace PowerShell;

- It provides a set of read/write file tools written in Rust, returning compact, line-numbered results with clearer error messages, so there are fewer back-and-forth rounds;

- It hides some official tools and loads heavy jobs like reading PDFs and Word files on demand, so the system prompt doesn't balloon.

The early version's Rust tools directly used code from FastCtx, an open-source project already validated in the Codex community (Apache-2.0 license), and were only replaced with the author's own code after 0.3.2. Put plainly, it carries the token-saving ideas from Claude Code and Codex (rtk has 85k stars, FastCtx has 1530 stars) over to DSH, and then specifically handles Windows.

### Where the stars came from

**The author's own Bilibili account is the main channel.** "魔法师風雪" has only 4989 followers, but he's not new: in 2025 July he posted an open-source galgame video that got 125k plays, and from 8 to September he posted several DSH-related videos (61.9k and 46.5k plays). The viewers who follow DSH were already on his channel. The dsh-ops collection has 4 videos in total: the 10-09 one has 250k, the 10-14 one 384k, the 10-21 version update 54k, and a 10-26 video on DeepSeek V4.1 Flash's "hidden features" 67k, about 750k plays combined. Both star surges match the video release dates.

**A small cross-promoting crew.** The README names ccch1mneyyy, the dsh-TUI author, as a maintainer. dsh-TUI has 6990 stars, and in his video the author calls it "the TUI recommended by DeepSeek's official product lead." The README also has a section titled "Linked!" listing dsh-TUI, DeepSeek-Balance-Whale-Widget, dsh-tauri and niubash, ending with "Remember to star them~." The desktop client dsh-tauri bundles dsh-ops directly, so people who install the client get it too. The author also leads two community spec repos (dsh-std, dsh-ecosystem-spec) and has been active in the DSH community since August.

**Word of mouth keeps spreading.** On linux.do, 10-22, someone posted recommending must-install plugins, and the line about dsh-ops reads: "Not to be overlooked, a blessing for Windows users... ideal for pure Windows setups without WSL." On X, game backend lead @cholf5 reposted the "27x" video on 10-17. On CSDN, someone wrote a troubleshooting article, "Four classes of dsh-ops install and run failures."

I found no sign of ads, Hacker News, Reddit or English-language media coverage. Its reach is limited to this small circle of Chinese DeepSeek users.

### Are the people who installed it using it?

Yes. npm downloads peaked about a week after the star peak: in October it was 2–3 thousand a day, rising from 10-22 to 10-28 to 8 thousand to 13k a day (peak 10-28, 12,872 downloads), then falling back to 5.5–6.5 thousand a day in November. Stars come from people watching the videos and clicking; downloads come from people who actually installed it and keep updating. But I can't tell how many come from dsh-tauri bundling versus auto-updates (estimate).

Star growth is already cooling: 523 in the last 7 days, 937 in the 7 days before that, and now a steady 60–85 a day. Forks are only 3.5% of stars, while similar repos of the same age and size typically sit around 11%. That suggests most people starring it are users installing the plugin, not developers modifying the code.

Two things to discount. First, "27x" and "saves 40%" are the author's own measurements. Someone on Bilibili ran a controlled comparison: simple tasks actually cost about 8% more, while complex tasks saved 13.5% and ran 33% faster. The effect is real, just not as dramatic as the titles claim. Second, the 6 release on Nov 0.5.0 has a serious bug: the cache isn't cleaned up, so disk usage keeps growing. The top of the README says "Please disable it ASAP, awaiting a fix." How much this hurts retention is still to be seen.

### Who it's competing with

"Saving tokens" in DSH is not a new category. Dozens of plugins were working on it as early as August, and dsh-ops isn't the first:

| Project | What it does | Scale (11-06) | How dsh-ops differs |
|---|---|---|---|
| DSH official compression | Context compression | On by default | Users complain "it's really been pulled" |
| dsh-context | Context visualization and management | 4245★, npm 30 days 380k downloads | No tool swap, ignores Windows |
| billion-context | Context compression, claims 5x savings | 4111★ | Complementary; often recommended alongside dsh-ops |
| dsh-sieve | Filters tool output | 110★ | Closest direction, but no Windows bash and nobody made videos |
| dsh-plugin-save-token | Generic token saving | 69★ | No specific scenario |
| Off-peak savings plugins (at least 7) | Run it in off-peak hours | All small | It saves on price per token, not on volume |
| fauxnix | Translating bash into PowerShell | 898★ | Same Windows pain point, but not a DSH plugin |
| rtk | Compresses command output by 60–90% | 85,644★ | Cross-agent, not native to DSH |

Of the projects above, only dsh-ops picked the angle of 'tools + Windows shell' and reached several thousand stars. I haven't found any project that clearly copies it. The similarly named ones (like a Windows ops tool) just share the name.

### Why this one

- **The pain point is concrete and users can see it** : failed commands, retries, burned money. Everyone on Windows using DSH has hit it.

- **The selling point is a single number** : 'saves 40%', '27x'. That is far easier to pitch than 'context management'. DeepSeek users are already price-sensitive, and there are 7+ plugins that just save money via off-peak pricing. On Zhihu someone complained they 'burned nearly ¥50 in two days'.

- **It caught a platform window** : DSH went open source in August, and by October the repos tagged dsh-plugin had passed 10k. In October DeepSeek V4.1 Flash launched and brought in another wave of new users.

- **Distribution was ready-made** : a Bilibili account with an existing audience, a team that recommends plugins officially, and built-in placement in the client. Cold start barely began from zero.

- **The code was ready-made too** : starting from an implementation already validated in another ecosystem meant it worked on day one.

Differentiation made it memorable and recommendable (linux.do recommended it specifically for Windows). But the star climb came from the videos and the author's channels.

### If it were you

**Verdict: B.** The playbook can be copied, but this specific position is already taken. The author's existing Bilibili audience, the dsh-TUI team, and built-in client placement are three things a latecomer can't get in the short term. The DSH ecosystem has also gone from 'few plugins' to 'ten thousand repos'.

**Who it suits** : developers already using DSH (or another agent tool that just opened plugins) who can clearly name where a specific system or scenario hurts. Ideally they can cut a 1–3 minute short video themselves, or know a creator willing to co-post.

**Who it doesn't suit** : anyone building yet another generic 'save tokens / compress context' plugin. That slot already has a dozen-plus entries. Also anyone with no content channel who plans to just post the repo and wait to be discovered. The dsh-sieve direction is nearly the same, nobody made videos, and it stalled at 110 stars.

**How to do it** : don't head-on compete for Windows shell. Find a specific pain point it hasn't solved and nobody owns. Three directions: first, similar 'model keeps tripping up' scenarios on macOS/Linux; second, a class of heavy tasks such as spreadsheets, databases, or frontend debugging, where multiple rounds of errors get compressed into one; third, port tools that already have thousands or tens of thousands of stars in the Claude Code ecosystem (tools like rtk and serena) into native DSH plugins. You can start from open-source implementations (mind the licenses and credit the source), and it must be findable and installable in the plugin marketplace on launch day.

**How to get users** : copy its three steps. ① Put a measurable number in the title, pair it with a 1–3 minute Bilibili video, post it yourself or co-post, and link the repo in the description. ② Reply on linux.do threads like 'must-install DSH plugins'. Several of these were active in October, with views in the 1,000 to 2,000+ range. ③ Get the authors of adjacent plugins or clients to list each other in a 'works with' table in the README, and aim for built-in placement in the client. Have the numbers re-tested by a third party. dsh-ops's '27x' was already shown to be overstated in testing, and latecomers exaggerating will get caught more easily.

**When to launch** : the DSH plugin ecosystem is still rising, and each new DeepSeek model brings in new users. The best window is the two weeks before or after the next major release or new model, so you catch that wave looking for plugins. Once a 'must-install' list sets, new plugins struggle to break in (inference).

**What would show you read it wrong** : ① DSH ships its own Windows bash or a better toolset, which makes this category of plugin largely useless. ② After dsh-ops's 0.5.0 hard-disk bug, npm daily downloads fell below 2k, which would mean users aren't sticking, and demand on this path is thinner than it looks. ③ Conversely, if a second DSH plugin built on 'one specific pain point + one Bilibili video' reaches a thousand stars within a month, the path is still open, so move fast.

[GitHub repo ↗](https://github.com/T-Auto/dsh-ops)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| DSH official dsh-compaction-basic | Big-company features | 2026-08 | On by default | The official one only compresses context, and users complain it works poorly. dsh-ops changes the tool surface and the shell. |
| bowenliang123/dsh-context | Open source | 2026-08-14 | 4245★, npm 30 days 380k downloads | Context visualization and management. Doesn't swap tools or handle Windows shell. |
| ranxianglei/billion-context | Open source | 2026-08-06 | 4111★ | Context compression. Complementary to dsh-ops and often recommended alongside it. |
| Sev7eEn7/dsh-sieve | Open source | 2026-10-04 | 110★ | Filters tool output. Closest in direction, but no Windows bash and no video distribution. |
| vibe-any/dsh-plugin-save-token | Open source | 2026-08-27 | 69★, npm 30 days 17k downloads | Generic token-saving plugin with no clear scenario. |
| yc-duan/fastctx | Open source | 2026-07-17 | 1530★ | Codex's MCP tool. dsh-ops initially bundled its code directly, which brought it into DSH. |
| rtk-ai/rtk | Open source | 2026-01-22 | 85,644★ | Cross-agent command-output compression proxy. Not a native DSH plugin, and doesn't address Windows shell. |
| 20000419/fauxnix | Open source | 2026-08-16 | 898★ | MCP that translates bash into PowerShell. Solves the same Windows pain point, but isn't a DSH plugin. |

Where users come from

- **Verified** **Author's own Bilibili short videos** : the dsh-ops collection from 魔法师風雪 (4989 followers): 10-09 'One click saves the big fish 40% of tokens' with 251k views; 10-14 co-posted 'Optimizing 27x token consumption' with 384k views, matching the 10-15/16 star peak. The GitHub link is in the description.

- **Verified** **dsh-TUI team cross-promotion, client built-in** : a maintainer, ccch1mneyyy, is the author of dsh-TUI (6990★, officially recommended plugin). The README 'works with' tables ask each other for stars. The dsh-tauri desktop app bundles dsh-ops.

- **Verified** **X repost** : @cholf5 reposted the video and repo link 27x in 2026-10-17.

- **Verified** **Community word of mouth (linux.do / CSDN)** : a 10-22 must-install plugin thread on linux.do calls it 'a blessing for Windows users'. CSDN has troubleshooting articles.

- Inference **Install via DSH plugin marketplace search** : the README's main install method is searching the npm package name in the marketplace. The npm download peak came a week after the star peak, which suggests the marketplace and built-in placement drove installs.

## Why T-Auto/dsh-ops took off

> Data as of 2026-11-06. DSH = DeepSeek Harness, the agent harness open-sourced by DeepSeek 2026-08-13, built around 'everything is a plugin' (github deepseek-ai/deepseek-harness, 274k stars in the repo).

### One line

dsh-ops is DSH's 'token-saving low-level tool plugin': on Windows it replaces the official PowerShell with a bundled bash, adds a set of Rust-written file and document tools that return compact results with line numbers (early versions bundled code from the Codex plugin FastCtx directly), and can hide some official tools. It got 29 stars in 5298 days, mainly through the author himself (Bilibili UP host '魔法师風雪', also a dsh-TUI team member) posting Bilibili short videos in a row: two videos with 250k and 380k views. The pain point it hit was 'DeepSeek repeatedly types wrong PowerShell commands on Windows and burns tokens for nothing'.

### 1. Data (GitHub repo_profile / repo_day, npm)

- Repo created 2026-10-08; the npm package was published the same day (registry.npmjs.org/dsh-ops, created 2026-10-08T11:18Z, 17 versions in total).

- Star curve (GitHub repo_day; data before 10-27 from deeply): 10-10 50 → 10-12 348 → 10-13 656 → 10-15 1420 (+584) → **10-16 2219 (+799, single-day high)** → 10-20 3273 → 10-28 4621 → 11-06 5298. Last 7 days +523, and the 7 days before that +937. The status is 'cooling', but it still gains steadily, 60–85 per day. Forks 188, only 3.5% of stars, below the median of repos of similar age and size by about 11%. No suspicious flags.

- Real usage: npm downloads from 10-08 to 11-05 total about 144k (api.npmjs.org). Daily, it started at 2–3K a day from 10-10, then ran 8–13K a day from 10-22 to 10-28 (peak 10-28 12,872), and fell back to 5.5–6.5K a day in November. Downloads peaked a week after stars did, which suggests people are installing it and actually using and updating it. Same-period comparisons over 30 days: dsh-context 380k, billion-context 2.99M (likely pulled in as a dependency by other packages), dsh-plugin-save-token 17k, dsh-sieve 1584.

- The author account T-Auto (GitHub display name '风雪T-Auto') has two other repos: dsh-std (160★, 'DSH plugin interoperability meta-protocol') and dsh-ecosystem-spec (89★, a community spec jointly launched by the dsh-TUI team and the DSH-Desktop-EAC team). That shows he has been working on ecosystem protocols in the DSH community since August, so he is not a newcomer.

### 2. Competitors: who does the same thing (saves tokens for agents / swaps in better low-level tools)

| Name | Where | Launched | Scale | How dsh-ops differs |
|---|---|---|---|---|
| DSH official dsh-compaction-basic and official switches | Big-company features | Ships with DSH 2026-08 | On by default | Compresses context only. A linux.do user said 'the official context compression really drags'. dsh-ops doesn't touch compression and changes the tool surface instead. |
| bowenliang123/dsh-context | Open source (DSH plugin) | 2026-08-14 | 4245★, npm 30 days 380k downloads | Focuses on context visualization and management. Doesn't change the shell or the tools. |
| ranxianglei/billion-context | Open source (DSH plugin) | 2026-08-06 | 4111★ | Context compression claiming 5x token savings. Complementary to dsh-ops, and both get recommended together in linux.do threads. |
| Tyan66666/billion-context-dsh | Open source | 2026-08-13 | 168★ | Lets the model trim its own context (ACP) |
| Sev7eEn7/dsh-sieve | Open source | 2026-10-04 | 110★ | Filters tool output. Closest in direction, but no Windows shell solution and no video distribution. |
| vibe-any/dsh-plugin-save-token | Open source | 2026-08-27 | 69★, npm 30 days 17k downloads | Generic 'token-saving' plugins |
| TUcookie/dsh-offpeak-saver and other off-peak plugins (at least 7) | Open source | 2026-08 | Small | Saves money through off-peak billing, not by reducing the tokens themselves. |
| yc-duan/fastctx | Open source (Codex's MCP) | 2026-07-17 | 1530★, +16 in the last 7 days | dsh-ops initially bundled its Rust tools directly (Apache-2.0,0.3.2, removed later and replaced with in-house code). FastCtx serves Codex and never entered DSH. |
| rtk-ai/rtk | Open source (Claude Code, etc.) | 2026-01-22 | 85,644★ | Cross-agent CLI proxy that cuts command output by 60–90%. Not a native DSH plugin, and doesn't solve the Windows shell problem. |
| oraios/serena | Open source MCP servers | 2025-03 | 31k★ | Toolset for semantic search and editing. Similar direction, but heavy and not built specifically for DSH. |
| 20000419/fauxnix | Open source MCP servers | 2026-08-16 | 898★ | Translates bash into PowerShell so Linux commands run on Windows. Solves the same Windows shell pain point, but isn't a DSH plugin. |
| unixwin/niubash | Open source | 2026-03-28 | 226★ | Not a competitor but upstream: dsh-ops 0.5 ships its own Windows bash by default |

Bottom line: "save tokens" in DSH is already a crowded category; August alone had a dozen plugins. dsh-ops isn't the first. Its angle is the tool surface plus Windows shell, not context compression, and among DSH-native plugins it's the only one at several thousand stars on that angle. The Claude Code/Codex ecosystem has bigger equivalents (rtk 85k★). dsh-ops is essentially porting that approach (rtk/FastCtx) into DSH and specifically handling Windows.

### 3. Growth tactics: where users come from

1. **The author posted a video on Bilibili (main channel, confirmed)** . UP host "魔法师風雪" (space.bilibili.com/3461566903094104,4989 fans). The video description links the project at github.com/T-Auto/dsh-ops, and he also built a dsh-ops collection with 4 videos:
    - 10-09 "One-click plugin that cuts your big fish 40% token spend!", 251k views (BV1nFp46qECm). Posted the day after the repo was created; star and npm downloads started climbing 10-10 later.
    - 10-14 "Did you know your DeepSeek can cut token use by 27x? Eat less plain rice!" Co-produced, 384k views (BV1Ghpy6gE5x). **It lines up with the star peaks on 10-15 (+584) and 10-16 (+799).**
    - 10-21 "dsh-ops v0.4.1: faster, cheaper, more stable", 54.4k views; 10-26 a video on a hidden DeepSeek V4.1 Flash feature, 66.8k views.
    - A third-party UP also made a comparison test video (BV1wZxH6PEmP, 4560 views): on simple tasks +8% cost; on complex tasks −13.5% cost and −33% time. It helped build credibility but the reach was small.
    - The account had built an audience before this: 2025-07 an open-source galgame video with 125k views, and 8–September several dsh-TUI videos (61.9k and 46.5k views). So he already had viewers interested in DSH.

2. **The dsh-TUI team's cross-promotion network (confirmed)** . The README says the maintainers include ccch1mneyyy, the author of dsh-TUI. dsh-TUI has 6990★ and claims to be "the first officially recommended plugin in the DSH community"; the author's Bilibili video also calls it "the TUI recommended by DeepSeek's official product lead." The README has a "Cross-links!" section listing dsh-TUI (6k★), DeepSeek-Balance-Whale-Widget (6k★), dsh-tauri (3k★), and niubash, with the line "Remember to star them~". The dsh-tauri desktop client **bundles dsh-ops** , which is another source of installs.

3. **X reposts (confirmed, reach unknown)** . Zhou Erfu, @cholf5, reposted the 10-17 video and repo link at 27 (x.com/cholf5/status/2111440467783221448). He's a game backend lead who calls himself "30 billion tokens a day." This came after the peak and served to extend the momentum.

4. **Community word of mouth and plugin roundups (confirmed)** . A linux.do post by 10-22 lists it as a must-install: "Good news for Windows users… ideal for pure Windows setups without WSL." Another linux.do thread, "What are the must-have DSH plugins?", also mentions it as "a plugin for optimizing tokens." A CSDN author wrote a troubleshooting post, "Four types of dsh-ops install and run failures." It's also indexed by a SourceForge mirror, rising.dev, hdatf, and similar aggregator sites.

5. **DSH plugin marketplace search (inferred)** . You install it by entering the npm package name in the DSH plugin marketplace. npm downloads kept rising after the star peak, consistent with marketplace and client-bundled installs. I couldn't find marketplace ranking data.

6. No evidence found of ads, HN, Reddit, or English media coverage; searching the local x-radar corpus for "dsh-ops" returned 0.

### 4. Why it charted

- **Specific pain point you can picture** : many DSH users are on Windows. The official shell is PowerShell, but DeepSeek keeps writing bash-style commands, which fail, get retried, and burn tokens. dsh-ops offers a "works out of the box with bash" fix. A linux.do user's own words: "Good news for Windows users."

- **The selling point is a number** : "saves 40%", "27x optimization", paired with short Bilibili clips (as short as 26 seconds). That spreads far better than "context management." DeepSeek users are price-sensitive: at least 7 off-peak money-saving plugins launched around the same time, and on Zhihu someone complained of burning nearly ¥50 in two days.

- **Timing the platform window** : DSH went open-source in August, and by October the plugin ecosystem was taking off (over 10k repos tagged dsh-plugin). In October DeepSeek V4.1 Flash 1017 launched, and the author rode the hype (the README benchmark chart and the 10-26 video).

- **Ready-made distribution** : the author already ran a Bilibili channel with several thousand followers, is a member of the dsh-TUI team (an officially recommended plugin), and has the dsh-tauri client bundling it plus cross-links in READMEs. Cold start barely started from zero.

- **Built on proven implementations** : early versions shipped a Rust tool derived from FastCtx, already validated in the Codex scene; later it was replaced with an in-house version, and niubash was bundled in, so it worked on day one.

### 5. Differentiation

Most peers compress or manage context. dsh-ops changes the tool surface: it swaps out the official shell (on Windows it ships a full bash), hides some official tools, provides compact read/write tools with line numbers, gives PTC mode an SDK that returns JSON, and uses `tools_addition` to load heavy tools like PDF/Word on demand so the system prompt doesn't grow. It's one of the few DSH token plugins built specifically for **native Windows environments** . This differentiation **is why it gets remembered and recommended** (the linux.do recommendations are all about Windows), but **it isn't the direct cause of the star spike** : the Bilibili video's numeric hook drove the stars. Differentiation decides whether users stay (npm downloads kept rising after the star peak); distribution decides whether it takes off.

### 6. How many people can see it

- It spread in small circles: four Bilibili videos total roughly 750k views, plus linux.do, Chinese DSH communities, and X alt-account reposts. A Zhihu August roundup titled "I dug through 1 repos" came out early and didn't include it; I found no coverage from major accounts or media. It's an open-source project with no revenue.

- Copycats: none clearly found. Name collisions include MiraculousGarfield/dsh-ops (a Windows ops tool, different scope), dragon43pp/dsh-ops-skill, dsh-plugin-ops, and 240xu/dsh-message-ops. These are name overlaps, not copies (creation dates couldn't be verified due to GitHub API rate limits).

### 7. Takeaways

- In new platform-style products (harnesses, plugin marketplaces), find a niche where one system is especially painful (here, the Windows shell), and port an approach already validated in another ecosystem (rtk/FastCtx).

- Run Bilibili ads with a single number ("saves X% / X times") plus a 20–30 second video, with the author on camera or co-producing. It can build traction in a day.

- Set up a small network of developers who cross-promote: a "cross-links" table in the README, mutual stars, and getting a client to bundle your plugin.

### 8. Uncertain

- "27x / 40%" is the author's own measurement. A third-party comparison showed about 13.5% savings on complex tasks, while simple tasks actually cost about 8% more.

- I didn't verify who co-produced the 10-14 video or how much traffic each side brought.

- Can't separate how many npm downloads came from dsh-tauri bundling or repeated updates.

- 0.5.0 (11-06) has a serious bug where the cache never gets cleared and disk use grows without limit. The README already tells users to disable it; the effect on retention is still unclear.

- Due to GitHub API rate limits, I didn't check stargazer details or the creation dates of same-name repos.

### Sources

- Repo README/CHANGELOG/PROVENANCE: [github.com/T-Auto/dsh-ops](https://github.com/T-Auto/dsh-ops)

- Our GitHub data (repo_profile, repo_day, owner), 2026-11-06

- npm:[registry.npmjs.org/dsh-ops](https://registry.npmjs.org/dsh-ops) ,[api.npmjs.org/downloads/range/2026-10-01:2026-11-05/dsh-ops](https://api.npmjs.org/downloads/range/2026-10-01:2026-11-05/dsh-ops)

- Bilibili: [space.bilibili.com/3461566903094104/upload/video](https://space.bilibili.com/3461566903094104/upload/video), BV1nFp46qECm, BV1Ghpy6gE5x, BV1Aoni6YECh, BV1wZxH6PEmP

- X:[x.com/cholf5/status/2111440467783221448](https://x.com/cholf5/status/2111440467783221448)(2026-10-17)

- linux.do:[linux.do/t/topic/3059945](https://linux.do/t/topic/3059945)(2026-10-22)

- Zhihu DSH ecosystem roundup: [zhuanlan.zhihu.com/p/2074660252738442171](https://zhuanlan.zhihu.com/p/2074660252738442171) (2026-08-23 verified)

- CSDN:[blog.csdn.net/weixin_63007863/article/details/168214950](https://blog.csdn.net/weixin_63007863/article/details/168214950)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.