Skip to content
Back to search

whathappened

Agent skill: adaptive X-only briefing of what happened + public opinion + debates

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/kunchenguid--whathappened/badge.svg)](/listings/kunchenguid--whathappened)

HTML

<a href="/listings/kunchenguid--whathappened"><img src="/api/listings/kunchenguid--whathappened/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

# install into Grok Build (global recommended)
$ npx skills add kunchenguid/whathappened -g
# or: copy skills/whathappened → ~/.grok/skills/whathappened

# in Grok Build
/whathappened Kimi K3

Example output from a real run on launch day (edited only for README length; structure is what the skill produces):

# /whathappened: Kimi K3

**Window:** last ~24h · mode Same-day · as of 2026-07-16 ~22:30 UTC
**X sample:** ~60+ posts · Top + Latest + Semantic + first-party · 4 threads · confidence high (event) / medium-high (opinion; day-0)
**Entity resolve:** none (X user search → official @Kimi_Moonshot)

## What happened

Moonshot’s Kimi account launched Kimi K3 today as “Open Frontier Intelligence”: a 2.8T-parameter, 1M-context, native multimodal model for long-horizon agentic coding, live on Kimi product surfaces and API, with full weights promised by July 27, 2026.

The official thread also stresses architecture (Kimi Delta Attention, Attention Residuals, sparse MoE) and demos (kernel optimization, vision-in-the-loop game/UI work).

Independent and community scoreboards moved quickly: Arena Frontend Code Arena #1 (over Claude Fable 5), Text Arena roughly top-10, and Artificial Analysis Intelligence Index ~57 (near Opus 4.8 / GPT-5.5, behind Fable 5 and GPT-5.6 Sol).

## Where the conversation is

- Dominant frame: open weights (or soon-to-be) at near-frontier quality, especially coding/frontend - often cast as a milestone for open models.
- Secondary frames: API price/latency vs closed flagships; China vs US competitive pressure; policy/safety backlash expected next.
- Who is loud: Moonshot first-party; eval accounts (Arena, Artificial Analysis); coding-agent vendors (e.g. Cline); AI Twitter builders; a thinner but clear skeptic lane on cost/speed/real-work gaps.

## Public opinion map

Shares are rough qualitative judgment from this X sample, not polling.

| Camp | Share (rough) | Core claim | Representative voices |
|------|---------------|------------|------------------------|
| Open-frontier bulls | ~50% | Near Fable/Sol on many axes; #1 frontend; open weights change the game | @cline, @arena, @TheAhmadOsman |
| Nuanced practitioners | ~25% | Very strong (esp. frontend), not undefeated overall; slow / thinks hard / API not “cheap open” | @goodworse, @DimitriGilbert, @runzhuotao, @jaminball |
| Geopolitics / policy | ~15% | “Sputnik” for the US; fear of open-model bans; cycle of China/safety rhetoric | @MTSlive (quoting Amp CEO), @Dan_Jeffries1, @haider1 |
| Maximal hype | ~10% | AGI-adjacent language, “mogs everything” | e.g. @dedene |

## The live debates

1. **Is it frontier, or “almost frontier”?**
   Side A: same ballpark as GPT-5.6 / Fable 5, #1 frontend. Side B: AA puts it behind Fable/Sol overall; some real-workload tests put it nearer Opus 4.8 / GPT-5.5.

2. **Cheap open win vs expensive reasoning tax**
   Side A: $3/$15 per MTok framed as Sonnet-like value. Side B: “thinks a lot → costs a lot,” slower than hoped, pricier than prior Kimi models.

3. **Competition catalyst vs regulatory panic**
   Side A: healthy race, progress accelerates. Side B: expect essays, China-hawk policy, open-weight restrictions.

4. **“Open” today vs weights on July 27**
   API/product is live; weights are scheduled, not yet downloadable.

## Notable posts (receipts)

- @Kimi_Moonshot - origin: official launch specs + July 27 weights
- @arena - scoreboard: Frontend Code Arena #1 over Fable 5
- @ArtificialAnlys - independent eval: AA ~57; agentic gains; cost/task framing
- @cline - builder frame: largest open-weight class model, near Fable/GPT-5.6, Sonnet-like API price
- @synthwavedd - receipt: “drops at 57 on Artificial Analysis, just behind Fable and Sol”
- @goodworse - steelman critique: frontend strength vs real-world closer to Opus 4.8/GPT-5.5; slow; pricier than K2.7
- @DimitriGilbert - hands-on: “really good… only downside, it thinks a LOT… and therefore costs a LOT”

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 230 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <h1 align="center">/whathappened</h1>
  2. <p align="center">
  3. <a href="https://docs.x.ai/build/overview"
  4. ><img
  5. alt="Grok Build only"
  6. src="https://img.shields.io/badge/Grok%20Build-only-black?style=flat-square"
  7. /></a>
  8. <a href="https://agentskills.io"
  9. ><img
  10. alt="Agent Skills"
  11. src="https://img.shields.io/badge/Agent%20Skills-package-blue?style=flat-square"
  12. /></a>
  13. <a href="LICENSE"
  14. ><img
  15. alt="License"
  16. src="https://img.shields.io/badge/license-MIT-green?style=flat-square"
  17. /></a>
  18. <a href="https://x.com/kunchenguid"
  19. ><img
  20. alt="X"
  21. src="https://img.shields.io/badge/X-@kunchenguid-black?style=flat-square"
  22. /></a>
  23. <a href="https://discord.gg/Wsy2NpnZDu"
  24. ><img
  25. alt="Discord"
  26. src="https://img.shields.io/discord/1439901831038763092?style=flat-square&label=discord"
  27. /></a>
  28. </p>
  29. <h3 align="center">What just happened on X - and what is the room actually saying?</h3>
  30. <p align="center"><strong>Grok Build only.</strong> This skill needs Grok’s native X tools. Other agents can install the package; they cannot run it for real.</p>
  31. Something drops. A model ships. A founder posts. A product melts down.
  32. You open X and get dunks, screenshots, quote-tweet chains, and three conflicting "official" takes - none of them in one place.
  33. **whathappened** is a [Grok Build](https://docs.x.ai/build/overview) [Agent Skill](https://agentskills.io) that turns that firehose into a short neutral briefing: what happened, where the conversation is, the opinion map, the live debates, and the receipts.
  34. - **X-first** - public conversation from X, not a blog roundup. One optional web lookup only to resolve *who/what* the topic is.
  35. - **Adaptive window** - breaking stories use minutes or hours; quieter ones widen only when the story is incomplete. Freshness wins by default.
  36. - **Opinion map + debates** - camps, rough sample shares, steelman vs critique, with real post links - not a vibes paragraph.
  37. ## Quick Start
  38. ```sh
  39. # install into Grok Build (global recommended)
  40. $ npx skills add kunchenguid/whathappened -g
  41. # or: copy skills/whathappened → ~/.grok/skills/whathappened
  42. # in Grok Build
  43. /whathappened Kimi K3
  44. ```
  45. Example output from a real run on launch day (edited only for README length; structure is what the skill produces):
  46. ```markdown
  47. # /whathappened: Kimi K3
  48. **Window:** last ~24h · mode Same-day · as of 2026-07-16 ~22:30 UTC
  49. **X sample:** ~60+ posts · Top + Latest + Semantic + first-party · 4 threads · confidence high (event) / medium-high (opinion; day-0)
  50. **Entity resolve:** none (X user search → official @Kimi_Moonshot)
  51. ## What happened
  52. Moonshot’s Kimi account launched Kimi K3 today as “Open Frontier Intelligence”: a 2.8T-parameter, 1M-context, native multimodal model for long-horizon agentic coding, live on Kimi product surfaces and API, with full weights promised by July 27, 2026.
  53. The official thread also stresses architecture (Kimi Delta Attention, Attention Residuals, sparse MoE) and demos (kernel optimization, vision-in-the-loop game/UI work).
  54. Independent and community scoreboards moved quickly: Arena Frontend Code Arena #1 (over Claude Fable 5), Text Arena roughly top-10, and Artificial Analysis Intelligence Index ~57 (near Opus 4.8 / GPT-5.5, behind Fable 5 and GPT-5.6 Sol).
  55. ## Where the conversation is
  56. - Dominant frame: open weights (or soon-to-be) at near-frontier quality, especially coding/frontend - often cast as a milestone for open models.
  57. - Secondary frames: API price/latency vs closed flagships; China vs US competitive pressure; policy/safety backlash expected next.
  58. - Who is loud: Moonshot first-party; eval accounts (Arena, Artificial Analysis); coding-agent vendors (e.g. Cline); AI Twitter builders; a thinner but clear skeptic lane on cost/speed/real-work gaps.
  59. ## Public opinion map
  60. Shares are rough qualitative judgment from this X sample, not polling.
  61. | Camp | Share (rough) | Core claim | Representative voices |
  62. |------|---------------|------------|------------------------|
  63. | Open-frontier bulls | ~50% | Near Fable/Sol on many axes; #1 frontend; open weights change the game | @cline, @arena, @TheAhmadOsman |
  64. | Nuanced practitioners | ~25% | Very strong (esp. frontend), not undefeated overall; slow / thinks hard / API not “cheap open” | @goodworse, @DimitriGilbert, @runzhuotao, @jaminball |
  65. | Geopolitics / policy | ~15% | “Sputnik” for the US; fear of open-model bans; cycle of China/safety rhetoric | @MTSlive (quoting Amp CEO), @Dan_Jeffries1, @haider1 |
  66. | Maximal hype | ~10% | AGI-adjacent language, “mogs everything” | e.g. @dedene |
  67. ## The live debates
  68. 1. **Is it frontier, or “almost frontier”?**
  69. Side A: same ballpark as GPT-5.6 / Fable 5, #1 frontend. Side B: AA puts it behind Fable/Sol overall; some real-workload tests put it nearer Opus 4.8 / GPT-5.5.
  70. 2. **Cheap open win vs expensive reasoning tax**
  71. Side A: $3/$15 per MTok framed as Sonnet-like value. Side B: “thinks a lot → costs a lot,” slower than hoped, pricier than prior Kimi models.
  72. 3. **Competition catalyst vs regulatory panic**
  73. Side A: healthy race, progress accelerates. Side B: expect essays, China-hawk policy, open-weight restrictions.
  74. 4. **“Open” today vs weights on July 27**
  75. API/product is live; weights are scheduled, not yet downloadable.
  76. ## Notable posts (receipts)
  77. - @Kimi_Moonshot - origin: official launch specs + July 27 weights
  78. - @arena - scoreboard: Frontend Code Arena #1 over Fable 5
  79. - @ArtificialAnlys - independent eval: AA ~57; agentic gains; cost/task framing
  80. - @cline - builder frame: largest open-weight class model, near Fable/GPT-5.6, Sonnet-like API price
  81. - @synthwavedd - receipt: “drops at 57 on Artificial Analysis, just behind Fable and Sol”
  82. - @goodworse - steelman critique: frontend strength vs real-world closer to Opus 4.8/GPT-5.5; slow; pricier than K2.7
  83. - @DimitriGilbert - hands-on: “really good… only downside, it thinks a LOT… and therefore costs a LOT”
  84. - @theo - ecosystem: practical “which harness?” question, not pure scoreboard talk
  85. ## Gaps / caveats
  86. - Day-0 recency bias: likes and quotes still forming; Top lane is announcement-heavy.
  87. - Weights not out yet: open-weight celebration is partly forward-looking (July 27).
  88. - English/AI-Twitter skew; capacity/noise on Latest.
  89. - What would change the read: multi-day coding failures, weights delayed, or closed labs shipping clear counters.
  90. ```
  91. ## Grok Build only
  92. This skill is **not** a multi-harness research product. It depends on Grok Build’s native X stack:
  93. - `x_keyword_search`
  94. - `x_semantic_search`
  95. - `x_thread_fetch`
  96. - `x_user_search`
  97. If those tools are missing, the skill should **refuse** rather than fake an X briefing from web search. Installing into Claude Code, Cursor, Codex, etc. will not give you a working run.
  98. ## Install
  99. **Global (recommended)** for Grok:
  100. ```sh
  101. npx skills add kunchenguid/whathappened -g
  102. ```
  103. **Project-local:**
  104. ```sh
  105. npx skills add kunchenguid/whathappened
  106. ```
  107. **Direct into Grok’s user skills dir:**
  108. ```sh
  109. git clone https://github.com/kunchenguid/whathappened.git
  110. cp -R whathappened/skills/whathappened ~/.grok/skills/whathappened
  111. # or: ln -s "$(pwd)/whathappened/skills/whathappened" ~/.grok/skills/whathappened
  112. ```
  113. Grok also discovers skills under `.agents/skills/`, so a normal `npx skills add -g` install often lands where Grok can see it. Prefer `~/.grok/skills/` when you want the Grok-native path explicitly.
  114. **List skills in this package without installing:**
  115. ```sh
  116. npx skills add kunchenguid/whathappened --list
  117. ```
  118. ## How It Works
  119. ```
  120. topic
  121. │
  122. ▼
  123. ┌─────────────────┐
  124. │ optional web x1 │ resolve entity only (name / handle)
  125. └────────┬────────┘
  126. ▼
  127. ┌─────────────────┐
  128. │ pulse (Latest) │ measure velocity → commit window mode
  129. └────────┬────────┘
  130. ▼
  131. ┌─────────────────┐
  132. │ search lattice │ Top · Latest · Semantic · from: · debate
  133. └────────┬────────┘
  134. ▼
  135. ┌─────────────────┐
  136. │ thread fetch │ origin · official · camps · steelman
  137. └────────┬────────┘
  138. ▼
  139. ┌─────────────────┐
  140. │ rank + cluster │ engagement × freshness × on-entity
  141. └────────┬────────┘
  142. ▼
  143. briefing
  144. ```
  145. - **Modes:** Breaking (hours) → Same-day → Story (days) → Background (longer only if needed) - shortest window that still explains the story.
  146. - **Soft web rule:** at most one lookup for identity. Sentiment stays X-only.
  147. - **No discovery mode in v1:** you must name a topic (`/whathappened` alone asks for one).
  148. ## Usage
  149. | Invoke | Example |
  150. | ------ | ------- |
  151. | Slash | `/whathappened Kimi K3` |
  152. | Natural language | "what are people on X saying about the OpenAI board drama" |
  153. The agent should auto-invoke when the description matches (launch reaction, X/Twitter sentiment, "what happened with…").
  154. ## Package layout
  155. ```
  156. skills/whathappened/
  157. SKILL.md # agent contract (pipeline + output template)
  158. references/
  159. query-patterns.md # X advanced search recipes
  160. failure-modes.md # thin sample, bots, entity collisions, …
  161. ```
  162. Ships as an [`npx skills`](https://github.com/vercel-labs/skills) package (`skills/<name>/SKILL.md`) for install convenience. Runtime target remains Grok Build only.
  163. ## Development
  164. ```sh
  165. # edit the skill
  166. $EDITOR skills/whathappened/SKILL.md
  167. # smoke-test discovery locally
  168. npx skills add ./ -l
  169. # install from this checkout for Grok
  170. npx skills add ./ -g -y
  171. # or symlink into ~/.grok/skills/whathappened
  172. ```

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.