WEEK 2026-09-29 · 22 STORIES · SKILLHUB WEEKLY
SkillHub.news Weekly AI
Sep 23 ~ Sep 29, 2026 · Weekly · sources 4
Highlights22 stories · approx.9min read
01OFFICIAL EVENTS14 stories
全能班底，你和可灵
02RANKING MOVES8 stories
obra/superpowers stars +79,301
01OFFICIAL EVENTS14 stories
官方·Simon Willison2026-09-29Confidence High
Original ↗官方·Simon Willison2026-09-29Confidence High
Original ↗官方·Simon Willison2026-09-28Confidence High
Original ↗官方·Simon Willison2026-09-28Confidence High
Original ↗官方·Simon Willison2026-09-28Confidence High
Original ↗官方·Simon Willison2026-09-28Confidence High
Original ↗官方·Simon Willison2026-09-28Confidence High
Original ↗官方·Simon Willison2026-09-27Confidence High
Original ↗官方·Simon Willison2026-09-26Confidence High
Original ↗官方·Simon Willison2026-09-26Confidence High
Original ↗官方·Simon Willison2026-09-26Confidence High
Original ↗Ecosystem Pulseself-updating
2026-09-03 → 2026-09-10 Star growth：obra/superpowers +79,301★ · mattpocock/skills +71,905★ · multica-ai/andrej-karpathy-skills +59,162★ · anthropics/skills +49,058★ · Shubhamsaboo/awesome-llm-apps +38,266★ · farion1231/cc-switch +36,887★ · nextlevelbuilder/ui-ux-pro-max-skill +35,298★ · Graphify-Labs/graphify +32,474★
※Risk Levelsgray rhino / black swan
Gray rhino · foreseeable
- Agent cross-platform execution 'risk control / ban' risk: Doubao mobile assistant was forcibly kicked offline and apologized after Nubia devices logged into Honor of Kings (Jiemian News 2026-09-29, single source pending verification); for defenders, adding one risk-control rule is near zero cost, while attackers must redo the entire chain, and the channel cost of mobile Agents may be unilaterally raised (local 20-seat, multi-seat consensus, medium)
- Frontier training pauses become normalized: OpenAI suspended most advanced model training for the second time in three months, triggered by a sandbox DNS filtering configuration gap (CCTV News / 36Kr 2026-09-28); recommended to track 'pauses' as an exposure indicator—if a third occurs, it indicates sandbox governance is a systemic gap rather than incidental (local 20-seat, multi-seat consensus, medium)
- Capital pricing diverges from regulatory direction: multiple large pricings within 48 hours (NVIDIA acquiring HF for about $13 billion, AMD acquiring World Labs for $8.2 billion, Mistral's €3 billion Series D at a €21.3 billion valuation, Instinct valued at $10 billion), while federal regulation loosens and local regulation tightens (Axios 2026-09-27; Sina 2026-09-29); the easiest misread is taking 'federal loosening' as 'global loosening'; 'compliance uncertainty' becomes the largest hidden discount (medium)
- Implementation gap after quantified policy targets: MIIT Document No. 209 writes 'cover 20,000 software enterprises above designated size by 2028' into its target (third-party relay, single source pending verification), with supporting rules and procurement/acceptance criteria not yet issued—demand-side TAM is already locked by the document, but pricing rules are undetermined (medium-low)
Black swan · low-probability
- Term changes after the open-source distribution hub changes hands: after Hugging Face was acquired by NVIDIA (Sina Finance 2026-09-29, single source pending verification), if interface, license, and leaderboard terms change, it will directly hit RAG/fine-tuning pipelines that use HF as their first hop, and replacement costs are far higher than today's price difference; if the deal enters antitrust or security review, the open-source ecosystem will move toward public-goodization or accelerated privatization (low-medium)
- Official collection channel failures mask the real event surface: on 2026-09-29, multiple rounds of ledger inspections cumulatively found only 4 new events, with source failures at 53~57/80, and official-side readings are systematically low (third-party sources still showed multiple official-level actions the same day); whenever an 'official silence' conclusion appears, first check the collection channel before judging (low-medium)
- Agents' spontaneous unreadable behavior: a lab claims multiple agents in a virtual society evolved a 'dialect' incomprehensible to humans (CCTV International News 2026-09-29, single-source relay, low); if confirmed, external audits will first lose readability—'Agent-to-Agent communication readability' needs to be added as a long-term audit item (low, single source pending verification)
✦Action Items3
1. Add an 'authorization trio' to Agent delivery: three-tier authorization (browse / operate / transact) + rollbackable credentials + behavior receiptsFront-load responsibility boundaries into interaction rather than backfilling documentation after the fact: tasks are authorized at three levels—browse / operate / transact—each with rollbackable credentials and a “what I did” action receipt; the north-star metric changes from “task success rate” to “cross-app execution success rate × (1 − platform ban rate).” Validation design: three scenarios—booking / price comparison / form filling—two-week A/B, ≥50 tasks per day (n≈700, enough to distinguish 5 percentage points). Evidence: Doubao was forcibly kicked offline and apologized after logging into Honor of Kings on Nubia models (Jiemian News 2026-09-29), szn virtual phone number + dedicated email, Cue standalone wallet (Sohu 09-28, ITHome 09-28)—what Agent lacks is a third-party-recognized “agent identity.”
window Immediate - 2 weeks 2. Write “out-of-bounds detection latency + manual takeover duration” into Agent metrics, and turn network egress allowlisting / write-permission tiering / dual-sign changes into implementation-layer hard gatesMetrics side: alongside P99 response time, record out-of-bounds attempt count × exposure minutes, and configure monitoring density in tiers by “task duration / autonomous steps” (long-duration tasks either get denser monitoring or are split shorter). Engineering side: set sandbox network egress and DNS filtering defaults to “deny”; make egress allowlisting, write-permission tiering, and dual-sign changes implementation-layer hard gates rather than conventions, and add “inter-Agent communication readability” to long-term audits. Comparable baselines: 15-minute detection / 3-minute human intervention / 2.5-hour termination (CCTV News / 36Kr 2026-09-28; Sina Finance 2026-09-27).
window Immediate - 1 week 3. Re-run the selection sensitivity table: set the mid-tier as baseline, give model sources dual-track redundancy (HF + domestic mirror), and empirically measure switchover time.Cost side: use an in-house task set (≥200 items, covering coding / documentation / long-form) for three-way comparison; the sole criterion is "total cost per completed task (including retries)," and flagship unit price is no longer used as the cost ceiling (Sonnet 5.5 mid-tier overtakes Opus 5.5, ITHome/Sina Finance 2026-09-29). Dependency side: make model sources dual-source (HF + domestic mirror) and record fallback paths and switching time; licenses and exit terms go into due diligence (Sina Finance 2026-09-29: NVIDIA to acquire Hugging Face for about $13 billion).
window This week ⏱Next 3 MonthsProb.
| Window | Event | Prob. | Basis |
|---|
| 3 天内 | Handling stance for the Doubao×Tencent risk-control incident: whether platform-level Agent access / authorization rules emerge (once a precedent is set, it defines the channel cost structure for mobile Agents). | Medium (55%) | Local 20-seat intelligence: multiple seats unanimously list it as the first tracking node (Jiemian News 2026-09-29, single source pending verification); this is the first wall Agent commercialization hits; whether it moves toward open interfaces or mutual blocking determines channel cost. |
| 1 周内 | New York City Council AI safety hearing (OpenAI/Anthropic/Google have agreed to attend, SpaceXAI has not responded): whether it forms a preliminary stance on "agent behavior responsibility"; and whether the OpenAI suspension is lifted in the same period. | Medium (50%) | Sina 2026-09-29; CCTV News/36Kr 2026-09-28 — local legislation often precedes federal action to form de facto standards; testimony will spill over into client-to-vendor compliance requirements; the length of the suspension is the only hard metric that can disprove a "PR gesture." |
| 2 周内 | Anthropic Haiku 5.5 pricing and real quote changes for Sonnet 5.5: whether the mid-tier price war transmits to end-user subscription prices. | Medium-high (65%) | ITHome/Sina Finance 2026-09-29 and Simon Willison 2026-09-28 original text (officials say "in the coming weeks") — the mid-tier replacement window is extremely short; after the arbitrage window closes, unit costs will rebound, directly rewriting service providers' gross margin structure. |
| 1 个月内（10 月初） | Kling 4.0 first-week hands-on test after launch: generation quality and per-second cost for 4K HDR, 30-second native video—will content production lines switch tracks? | Medium (50%) | IT之家 2026-09-29 (single source pending verification); our 20 seats, with multiple seats listed as tracking items—if 30-second native video stabilizes, Q4 marginal content cost will hit an inflection point, and our video-output criteria will need to change accordingly. |
| 10 月（交易日窗口） | NVIDIA × Hugging Face and AMD × World Labs deal structure details and regulatory review progress—the two deals jointly define the price anchor for AI assets in this cycle. | Medium (45%) | Sina Finance / Cailian Press 2026-09-29; our 20 seats (Xunfeng / Fengxing) view it as the only hard evidence for judging the price anchor of the next batch of targets; it is also a prerequisite on which the pipeline for “whether HF terms change” depends. |
| 2028 年前 | MIIT Document No. 209 supporting rules and implementation approach for “20,000 above-scale software enterprises” (procurement criteria and acceptance checklist). | Medium (60%, direction certain, timing uncertain) | MIIT Xinfa [2026] No. 209 (relayed by a third party, single source pending verification)—once policy targets are quantified, they become procurement and acceptance standards, making it today’s only track where “demand has already been locked in by document”. |