# An "never truncate context" 837-line agent: uncut hit 5 days, 611 stars, most of them before anyone knew it existed · GitHub open-source project

> Source: https://stringers.io/en/github/gh_Balanov__uncut.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

> It was born with about 360 stars (inferred to be seeded), cleared the new-repo rising threshold quickly, and was picked up by Trendshift's Go chart at #3 and a Chinese new-repo chart. The "no truncation" pitch was only recognized by a small circle.

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

Balanov/uncut · Balanov

# An "never truncate context" 837-line agent: uncut hit 5 days, 611 stars, most of them before anyone knew it existed

**What it is** A 7MB single-file Go command-line tool that lets an LLM read and write code and run commands. It guarantees the full conversation is sent every time, and errors out if output gets cut off.

**Why it blew up** It was born with about 360 stars (likely seeded), which got it past the new-repo star-growth thresholds in one shot. Trendshift ranked it #3 on the Go chart and Chinese new-repo lists picked it up. The "no truncation" pitch only resonated with a small circle.

**C** **How you can follow it** Don't build yet another minimal harness. Copy its positioning method: in a category where everyone keeps adding features, subtract, put "what this doesn't do" on the first screen, pair it with a reproducible comparison that names the top projects, then post your own Show HN and submit PRs to awesome lists.

Grade C: the 600 stars mostly come from stars it was born with. The pitch itself got under 1500 views. In the minimal-harness space, Pi has 120k stars and mini-swe-agent has 9k, so copying it won't easily get the same stars. What you can learn is the positioning writing. Few people know about it yet

Data (updated daily) · repo created 2026-11-01 · 255 stars

GitHub · stars

Data 2026-11-03 → 2026-11-08 (6 days, 6 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

uncut is a coding agent shell written in Go that runs only in the command line, and its pitch is that not a single byte of context is cut. Built 5 days ago, it has 611 stars and looks like a new breakout. But lining up the timeline, about 360 of those stars were there before the first post introducing it. The same author's 10 repos created in 8 days show the same pattern. This piece covers what it is, where the stars came from, and what you can copy from it for getting stars on your own open-source project, and what you shouldn't.

### The first post had only 30 views.

On Nov 1 2026 17:19 (UTC), GitHub user Balanov created the repo uncut. Early the next morning, 04:58, an X account with 298 followers, @RepoGems, posted the first introduction. That post has only 30 views so far.

The problem is that by then it already had about 360 stars. Trendshift (a site that ranks GitHub repos by daily new stars) first saw it on Nov 1, recording +0 new stars. Over the following six days it added +247 in total (11-02 got +207 on the day and hit #3 on the Go daily chart, followed by +20, +9, +8, +3). GitHub's total on Nov 6 was 611, so about 364 stars were unaccounted for in between. They were already there when Trendshift first captured it. No public spread explains these stars.

### What it is

The README's first line reads "The lossless, headless coding agent harness": a lossless, headless coding agent harness. It's a 7MB static binary with 4 tools (read, write, edit, run commands), reads from stdin and writes to stdout, depends on no third-party Go libraries, and the implementation is 837 lines in total. It works with any OpenAI-compatible endpoint, including OpenRouter and a local Ollama.

Its pitch is entirely about what it doesn't do:

- **No context compression.** Every model call sends the entire conversation as is. File contents and command output are never truncated.

- **Doesn't count half-finished answers as success.** If the model's output hits the length limit and gets cut, it errors out and exits with a non-zero code. The repo includes a byte-level proof script you can run yourself (bench/proof.py).

- **No UI, telemetry, plugins, or automatic retries.** It's built for pipes and CI only: `git diff | uncut "Review this diff"`, and adding `--json` outputs an NDJSON event stream.

The landing page at balanov.pro/uncut benchmarks directly against Pi, which has 120k stars: startup 9.5ms vs 243ms, and size 6.8MB vs 107MB for node_modules.

This hits a real pain point. On X, @davidpereIsHIM put it this way: "half of harness pain is not knowing what got dropped from context". Half the pain of building agents is not knowing what was thrown out of the context. The Chinese curation account @vintcessun (恒星sun, 9k followers) also singled out its failure handling, where truncation means an error, for praise.

### Where the stars came from

Going through the author's profile makes things clear. The author is Anton Balanov, a Russian whose bio says he founded Balanov.PRO / BAIagent. He reports online that he was CIO/CPO at Sber, VTB, and VK. The account was registered 2017 years ago, but all of his 19 public repos were created after Oct 28, 2026:

| Library | Repo created | star(11-06) | fork | After birth |
|---|---|---|---|---|
| svoi (Windows job-search assistant) | 10-28 | 376 | 0 | Flat, still declining |
| baiagent-director | 10-28 | 306 | 1 | Only +5 stars in 8 days |
| baiagent-free (a static download page) | 10-28 | 223 | 0 | Flat |
| **uncut** | 11-01 | 611 | 17 | About +207 at birth 364,11-02, then single digits per day |
| shim | 11-02 | 467 | 8 | +46 about 420,5 days in |
| skein | 11-02 | 266 | — | +7 about 259,5 days in |
| verdict / latch / equal | 11-02~05 | 243 / 231 / 155 | — | — |

(Star counts from GitHub, 11-06; trends from Trendshift and GitHub daily snapshots.)

A static download page with 223 stars and 0 forks, and not one more star since birth. This shape matches the pattern described in academic research on star fraud (the CMU StarScout paper, arXiv 2412.13459). The accounts that starred can't be checked one by one, so the "star-buying" call is only an **inference** , but a confident one.

uncut is the only one of these repos with real follow-on signal. Its growth comes in three layers:

1. **About 360 stars at birth** (inferred to be seeded), which got it past the various new-repo star thresholds from day one.

2. **The new-repo charts and aggregator sites picked it up** (confirmed). It's #3 on Trendshift Go's daily chart and #97 on DKFile's "GitHub Rising Repos" chart, plus rising-repo, HotRepo and the bitdoze 11-05 daily digests, along with accounts on Threads and LinkedIn that auto-post "Trending on GitHub". All of these sites scrape repos by star growth, so the +207 on 11-02 is very likely coming from this layer.

3. **X alt accounts and the Chinese developer circle** (confirmed, small volume). Roughly 11 posts, with views ranging from 3 to 628, adding up to under 1500. The biggest was from Geek Lite (@QingQ77), with 29k followers, at 628 views. Several posts restate "no silent cuts" from the README and number the points in a way that looks auto-generated. Still, most of the 17 forks came from Chinese accounts, including the Go blogger smallnest (5k followers), so that attention is real.

The author also forked 6 awesome lists to submit PRs, and uncut has already been added to kyrolabs/awesome-agents. The landing page link carries utm_source=theweekly.com, which indicates a newsletter feature; which one, and whether it was paid, is unverified. It hasn't been on HN or Reddit, and no media has covered it.

### Who it's competing with

The "minimal agent shell" niche is already taken:

| Project | star | How uncut differs |
|---|---|---|
| Pi(earendil-works/pi) | About 120k, +1,506 in a week | Equally minimal, equally 4 tools, has a headless mode; written in TS, and doesn't promise no compression |
| mini-swe-agent | 9.1k | 100 lines, bash only, linear history with no compression, built for CI and evals in the first place. The philosophy is nearly identical |
| nanocode | 2.8k | Single-file Python, more interactive |
| Claude Code `-p` / Codex `exec` / Gemini CLI | each 100k+ | Both can run headless, but both auto-compress and tie to their own models |
| headroom / caveman / context-mode | 30k–120k | Opposite direction: compressing context to save tokens |

uncut has no new features. What's new is the framing: making "what I don't do" the selling point, plus a single Go binary. The author posted shim (467 stars) the next day, a variant of the same idea with MCP and Skills added. No outside imitators have shown up yet.

### If it were you

**One-line verdict: C.** The "no truncation" pitch resonates with some people, but the 600 stars mostly come from the batch present at birth, and the pitch itself got under 1500 views on X. Building a Go or Rust version of the same minimal shell, with Pi at 120k stars and mini-swe-agent at 9k blocking the way, won't reach a few hundred stars on the strength of the work alone.

**Who it's for:** Developers already building agent tools who have a real pain point and want to learn how to write positioning. **Who it's not for:** Anyone hoping to climb the stars by building another harness, or copying its cold-start playbook. Repos that seed stars show fork/star ratios so low you can spot them at a glance, and people who know the space will mark you down for it.

**Only two things are worth copying:**

1. **An anti-feature list plus a reproducible comparison.** While everyone is building compression and UIs, put "no compression, no retries, no UI" on the first screen of the README, attach a proof script anyone can run, and name the top projects directly with numbers (25x faster startup, 16x smaller). This move works elsewhere too: in any category where everyone keeps adding features, be the one that subtracts and can prove it.

2. **Small enough to read in full.** "837 lines, forkable" is itself a spreadable line that aggregators and curators can repeat in one sentence.

**How to get users (the clean version):** Do the half of the playbook that was real. Post Show HN and the relevant subreddits yourself on day one (uncut skipped both, which is a gap). Submit PRs to 3–5 relevant awesome lists. In Chinese-language circles, find curators in Go and agent topics. uncut's real forks mostly came from there.

**When to wait:** If you're not in a hurry. The "no context compression" message will only land widely after a major vendor's auto-compression causes a notable failure, and that timing can't be scheduled.

**What would show this read is wrong:** If uncut keeps adding dozens of stars a day for the next two to three weeks without seeding, or if the HN crowd picks up the "no truncation" debate, then the pitch carries more weight than its view counts suggest, and the grade should move up to B.

[GitHub repo ↗](https://github.com/Balanov/uncut)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| Pi (earendil-works/pi) | Open source | 2025-08 | About 120k stars, +7 in 1506 days | Equally minimal, equally 4 tools, with print/JSON modes. uncut is a single Go binary, headless only, and explicitly does no compression. Its landing page benchmarks startup speed and size directly against Pi |
| mini-swe-agent | Open source | 2025-06 | 9.1k star | 100 lines, bash only, linear history with no compression. The philosophy overlaps heavily. uncut swaps in a Go static binary with an OpenAI-compatible interface, and adds error-on-truncation |
| nanocode | Open source | 2026-01 | 2.8k star | A single-file Python Claude Code alternative, interactive-leaning; uncut is headless-only, JSONL sessions, and promises no truncation |
| Claude Code -p / Codex exec / Gemini CLI non-interactive mode | Big-company features | 2025 | each with 100k+ stars, the largest user bases | All can run headless, but they ship auto-compaction and are tied to their own models; uncut is lossless against any OpenAI-compatible endpoint |
| opencode / crush / aider / goose | Open source | 2023-2025 | 30k to 220k stars | A full TUI product with compaction and plugins; uncut deliberately ships none of that |
| headroom / caveman / context-mode | Open source | 2026 | 30k to 120k stars | Their pitch is compacting context to save tokens; uncut's is the opposite: no compaction at all |

Where users come from

- Inferred **stars seeded at birth (bought or mutually exchanged)** : about 364 stars within ~11 hours of the repo being created, before the first external write-up. The same author's svoi (376 stars, 0 forks), baiagent-free (static download page, 223 stars, 0 forks), baiagent-director, skein, and shim were all born with two to four hundred-plus stars and have barely grown since (Trendshift and the github database)

- **Confirmed** **new-repo charts and aggregator sites** : Trendshift Go daily chart #3 (Nov 2, +207 that day); DKFile new-repo chart #97; rising-repo, HotRepo, bitdoze 11-05 daily digest, fluxreport, agentskillshub, plus the auto-generated "Trending on GitHub" posts on Threads and LinkedIn

- **Confirmed** **X burner and curator accounts** : about 11 posts, with views ranging from 3 to 628, totaling under 1500; several are AI-style posts that restate the README; the Chinese accounts Geek Lite and 恒星sun brought in Chinese-speaking developers

- **Confirmed** **awesome-list PRs** : the author forked 6 awesome lists, and UNCUT has been listed in kyrolabs/awesome-agents

- Inferred **newsletter** : the landing page carries utm_source=theweekly.com&utm_campaign=weekly_discovery. Which newsletter it is, and whether it was paid, is unverified

## Balanov/uncut: a 837-line Go headless coding agent harness that "never cuts," and why it hit 600 stars in 5 days

> Research date 2026-11-06. Sources: GitHub API (live query), our github and hackernews databases, Trendshift pages, X (fxtwitter for original posts), Jina web search.

### Bottom line

Most of uncut's 600 stars were not organic. Within the first few hours after creation it picked up about 360 stars, before any write-up existed online. The same author created 10 repos in 8 days with the same "tiny / local-first / MIT / zero deps" pitch. Each one was born with two to four hundred-plus stars, has barely grown since, and has forks close to 0. The only organic traffic was part of the Nov 2 spike (+207), plus a tail of single-digit daily stars afterward. The sources were Trendshift's Go chart (#3), Chinese GitHub new-repo charts, and a dozen burner X posts. The product does have a clear contrarian hook (no compaction, no truncation, it errors instead of cutting), using Pi as its benchmark. But that hook drew under 2000 views on X, which cannot explain 600 stars.

### 1. Basic data

| items | value | Source |
|---|---|---|
| Repo created | 2026-11-01 17:19 UTC | GitHub API |
| star / fork | 611 / 17(11-06) | GitHub API |
| Daily stars | 11-03 570 → 11-04 589 → 11-05 600 → 11-06 604 | github repo_day |
| Trendshift daily | Nov 1 +0, Nov 2 +207 (Go chart #3), Nov 3 +20, Nov 4 +9, Nov 5 +8, Nov 6 +3; 6 days total +247 | trendshift.io/repositories/332152 |
| Commits | 7, the author wrote 6 of them themselves; code clustered between 11-01 and 11-02; the only external PR came from A2Agent (an API gateway vendor) | GitHub API |
| HN | None (the database has no story with ≥5 points, and there was no Show HN) | hackernews database |
| Reddit | No related posts found | Jina search |

Key point: Trendshift recorded 6 days of growth as +247, with a current total of 611, so roughly 364 stars are unaccounted for. Trendshift's Nov 1 row shows +0, which means that when it first saw the repo, it already had about 360 stars. Creation to the first external write-up (X @RepoGems, 11-02 04:58 UTC) was only about 11 hours. These stars did not come from any public promotion.

### 2. The same author's "born with stars" pattern (the most important evidence)

The author is Антон Николаевич Баланов (Anton Balanov), from Russia. His profile reads "Founder of Balanov.PRO / BAIagent." He claims online to have been CIO/CPO at Sber, VTB, and VK, and to teach as an IT-MBA professor (e-xecutive.ru, balanovanton.ru). His GitHub account was registered 2017 years ago, but all 19 of his public repos were created after 2026-10-28:

| Library | Repo created | star | fork | Post-birth trajectory |
|---|---|---|---|---|
| baiagent-director | 10-28 | 306 | 1 | Trendshift 8 days only +5; in our database it has been flat at 305 since 11-01 |
| svoi (Windows job-search assistant) | 10-28 | 376 | **0** | Flat at 377 since 11-01, still declining |
| baiagent-free (**a static download page** ) | 10-28 | 223 | **0** | 224 since 11-01, flat |
| balanov-pro-enterprise (Russian) | 10-31 | 59 | – | – |
| baiagent-distribution | 11-01 | 55 | – | – |
| **uncut** | 11-01 | 611 | 17 | Born at about 364 → Nov 2 +207 → single-digit daily gains |
| shim | 11-02 | 467 | 8 | Trendshift 5 days +46, meaning about 420 at birth |
| verdict | 11-02 | 243 | – | – |
| skein | 11-02 | 266 | – | Trendshift 5 days +7, meaning about 259 at birth |
| latch / equal / until / 2aos-core | 11-04~05 | 231 / 155 / 33 / 22 | – | – |
| 6 forked awesome lists | 11-01~02 | 1–2 each | – | Used to submit PRs for its own repos and get them listed (kyrolabs/awesome-agents has already listed UNCUT) |

(Star counts from the GitHub API 11-06; trajectories from the github repo_day data and Trendshift.)

One static download page picked up 223 stars and 0 forks, and a Windows job-search assistant got 376 stars and 0 forks, and neither grew after birth. This matches the buying and mutual-starring patterns described in the CMU StarScout paper (arxiv 2412.13459). Note that GitHub's stargazer lists were not accessible with our token in 2026 (returning 403/404), and GH Archive data for 2026 is heavily sampled (about 5 thousand events per hour), so we could not check the starring accounts one by one. The star-farming call rests on distribution patterns. It is a strong inference, not account-level evidence.

uncut is the only one in this batch with real follow-on signal: 17 forks (mostly Chinese accounts, including the Go blogger smallnest with 5k followers), +207 in a single day on Nov 2, and #3 on Trendshift's Go chart.

### 3. Where users came from (growth tactics)

1. **Born with stars (about 360, strongly inferred to be seeded)** . Evidence above: the timing precedes any public promotion, and the same author's other repos follow the identical pattern. The effect is that the repo qualifies for the various "new repos gaining stars" charts from day one.

2. **New-repo charts and aggregators (confirmed)** . Trendshift Go daily chart #3 (Nov 2); DKFile "GitHub new-repo chart" #97; yanggggjie's rising-repo; HotRepo, agentskillshub, fasttoolhub, trendar, aiagentideas, bitdoze 11-05 daily digest, fluxreport, TrendHub; auto-generated "Trending on GitHub" posts on Threads and LinkedIn (@arthi__rajendran, posted at ⭐576). These sites pick up repos based on star gains, and step 1 feeds them directly. The Nov 2 +207 most likely came from this layer. Still, only 5 new forks came in that day, and it's unclear how many of them were real people.

3. **Burner X posts (confirmed, small volume)** . Trendshift aggregated about 11 of them, and we read the originals one by one: @RepoGems (298 followers, 30 views), @arshadbarves (43 followers, 68), @LLMpsycho (1.5k followers, 88), @VoltableCom (104 followers, 12), @li90549 (72 followers, 17), @_aashish_singh_ (62 followers, 3), @davidpereIsHIM (787 followers, 39), @itssaadmemon (7 followers, 17), the Chinese accounts @vintcessun 恒星sun (9k followers, 174) and @QingQ77 Geek Lite (29k followers, 628). Combined, they drew under 1500 views. Many posts follow one template: they restate the README's "no silent cuts," use numbered bullet points, and come from accounts with few followers, so they look like AI-written auto-curation or mutual-promotion accounts. None of them appear in the x-radar local corpus.

4. **Chinese developer circles (confirmed, brought in real forks)** . Most fork accounts have Chinese names (smallnest, caishuai, dadaowubian, arwei944, etc.), and there were 3 Chinese X posts. This shows that Chinese new-repo charts and curator accounts drew some real attention.

5. **Awesome-list PRs (confirmed)** . The author forked 6 awesome lists, and UNCUT has already been merged into kyrolabs/awesome-agents.

6. **Landing page carries utm** . balanov.pro/uncut/?utm_source=theweekly.com&utm_campaign=weekly_discovery indicates a newsletter feature or ad placement. Which newsletter it was is unverified.

7. Sister repo latch was posted by @RoundtableSpace (Mario Nawfal's AI/crypto account, 270k followers): 44k views, 33 likes (11-04). This shows that some repos in the author's series reached big accounts, but uncut is not one of them.

No HN, no Reddit, no press coverage.

### 4. Competitors

| Product | Form | Launched | Scale | How uncut differs from it |
|---|---|---|---|---|
| Pi(earendil-works/pi,Mario Zechner) | Open-source TS | 2025-08 | 120k stars, +1506 over 7 days | Also minimal, also 4 tools (read/write/edit/bash), with print/JSON modes. uncut benchmarks directly against it: 9.5ms vs 243ms startup, 6.8MB vs 107MB node_modules. The difference is that uncut is a single Go binary, headless-only, and explicitly does no compaction. |
| mini-swe-agent | Open-source Python | 2025-06 | 9.1k | A "100-line agent" with bash as its only tool, and no compaction on its linear history. It was built for CI and benchmarks. uncut's philosophy overlaps heavily with it, but uncut ships as a Go binary and targets OpenAI-compatible APIs. |
| nanocode | Open-source Python | 2026-01 | 2.8k | A single-file 250-line Claude Code alternative. uncut adds headless mode, JSONL sessions, and the no-truncation guarantee. |
| learn-claude-code | Open source | 2025-06 | 80k | A teaching-oriented nano harness. uncut is a tool you can actually use, not a tutorial. |
| Claude Code `-p` / Codex `exec` / Gemini CLI non-interactive | Big-company features | 2025 | each with 100k+ stars, the largest user bases | All of these can run headless, but all ship auto-compaction and are tied to their own models. uncut's pitch is that it never silently cuts, against any OpenAI-compatible endpoint. |
| opencode / crush / aider / goose | Open source | 2023–25 | 3–220k | All are full products with a TUI, compaction, and plugins. uncut is their opposite. |
| headroom / caveman / context-mode | Open source | 2026 | 3–120k | Their pitch is "compact context to save tokens." uncut makes the opposite claim: no compaction at all. |
| Balanov/shim (same author) | Open-source Go | 2026-11-02 | 467 | Another 7MB Go harness from the same author, with MCP and Skills and interactive confirmation. It's basically copying its own homework into a variant. |

The landscape: the "minimal harness" niche is already taken by Pi and mini-swe-agent, and the leaders sit at six-figure star counts. uncut has no new features. What's new is the framing: "what it doesn't do" (no compaction / no retries / no TUI) as the selling point, plus a Go static binary.

### 5. Differentiation, and whether that's why it broke out

- Differentiation: (1) Lossless: the full conversation is sent every time, and file and command output is never truncated. If model output gets cut off, it exits with a non-zero code instead of passing off a half-finished answer as success. Ships a reproducible byte-level proof at bench/proof.py. (2) Headless only: stdout with NDJSON events, built for pipes and CI. (3) A Go static single file at 7MB, zero dependencies, 837 lines, easy to fork.

- This hits a real pain point in the harness crowd: not knowing what got cut from context. On X, @davidpereIsHIM put it as "half of harness pain is not knowing what got dropped from context". Chinese curation accounts also picked out "failure handling" for praise.

- But that isn't the main reason for the 600 stars. By the numbers: the roughly 360 stars it was born with came before anyone saw the pitch; after Nov 2 the organic tail is only 3–20 stars a day; total X views are under 1500. The pitch is recognized in a small circle, but the reach is tiny.

### 6. Visibility and copycats

- Visibility: almost nobody knows about it yet. The X posts are all small accounts under a thousand followers; the biggest is Geek Lite (29k followers, 628 views). No HN, no Reddit, no media. It has only shown up on aggregator sites and new-repo charts.

- Copycats: no outside copycats found. The author's own shim is a variant of the same idea. API gateway vendors like A2Agent watch repos that make new-repo charts and pitch PRs with their test keys, which suggests that charting alone can bring in commercial outreach.

### 7. What to borrow

1. Positioning through a "list of things it doesn't do": while everyone builds compaction and TUIs, write "I don't do X" as the selling point and back it with a reproducible benchmark, comparing directly against the leader (Pi) with 25x startup speed and 16x size difference.

2. Cold-start pipeline: seed stars at birth → clear the threshold of new-repo charts like Trendshift, DKFile, and rising-repo → get reposted by dozens of aggregators and auto-curation accounts → fork awesome lists and submit PRs → add utm tags and BibTeX citations to the landing page. The same template ships 8 repos a week, and the total stars across one account (2975) backs the author's BAIagent commercial brand. The cost is an extremely low fork-to-star ratio, which makes it easy to spot.

3. The small implementation (837 lines) is itself a talking point: "readable and forkable" appeals to developers.

### 8. Uncertain

- Whether the roughly 360 stars at birth were bought, traded among accounts, or concentrated from the author's private communities (Russian-language communities / Start Pervyh accelerator) — we can't get the stargazer list, so we can't check account by account.

- How much of the +207 in Nov 2 came from organic traffic.

- Which newsletter theweekly.com is, and whether it's a paid placement.

- The author's ultimate goal: most likely to give "international open-source" credibility to the BAIagent / Balanov.PRO B2B business in Russia. Not verified.

### Sources

- [github.com/Balanov/uncut](https://github.com/Balanov/uncut); GitHub API repos/Balanov/uncut, users/Balanov/repos, forks, issues, commits (2026-11-06 checked live)

- [trendshift.io/repositories/332152](https://trendshift.io/repositories/332152) (uncut);/329001(baiagent-director);/334744(shim);/334616(skein)

- [balanov.pro/uncut/](https://balanov.pro/uncut/) (landing page, benchmarks against Pi)

- Original X posts (read via api.fxtwitter.com): x.com/QingQ77/status/2118564280731861498, x.com/vintcessun/status/2117578544582967494, x.com/RepoGems/status/2117119178167451689, x.com/arshadbarves/status/2117155212142280761, x.com/LLMpsycho/status/2117169162002866438, x.com/davidpereIsHIM/status/2117972720470102411, x.com/RoundtableSpace/status/2118117449782460643 (latch)

- [dkfile.istester.com/bangdan/github/Balanov-uncut](https://dkfile.istester.com/bangdan/github/Balanov-uncut) ;[www.bitdoze.com/news/2026-11-05/](https://www.bitdoze.com/news/2026-11-05/) ;[yanggggjie.github.io/rising-repo/](https://yanggggjie.github.io/rising-repo/) ;[github.com/kyrolabs/awesome-agents](https://github.com/kyrolabs/awesome-agents)

- Author background: [www.e-xecutive.ru/users/1850976-anton-balanov](https://www.e-xecutive.ru/users/1850976-anton-balanov), [balanovanton.ru/](https://balanovanton.ru/)

- Our repo data: github.repo_profile / repo_day (Balanov/*); hackernews.story (no results)

- Star-inflation reference: [arxiv.org/html/2412.13459v2](https://arxiv.org/html/2412.13459v2)

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