nvidia-developer-firehose — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited nvidia-developer-firehose (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
NVIDIA's Technical Blog (developer.nvidia.com/blog) is where NVDA publicly names the partners it's designing the next generation around — 800V HVDC chip suppliers, CPO optical module makers, custom XPU/ASIC partners, networking/server OEMs, hyperscaler/neocloud customers. When a partner gets named here, it's typically weeks before the Substack / sell-side picks up on it. Forward-looking edge.
This firehose:
to auto-resolve every capitalized noun phrase → US-listed ticker.
nvdev_ticker_cache.json) learns over time sonewly-named partners get added automatically; never needs hand edits.
(high signal) from 🔍 newly discovered tickers (medium signal).
ATOM FEED CANDIDATE EXTRACT RESOLVE
───────── ───────────────── ───────
developer.nvidia.com → capitalized 1-4 word → cache HIT → done
/blog/feed/ phrases, minus cache MISS → yf.Search()
STOPWORDS + ALIAS name-overlap
expansion US exchange?
→ cache + emitNo hardcoded `name_to_ticker` dict. A small TRACKED_TICKERS set (your portfolio universe) only affects whether a hit shows up as 🎯 or 🔍. Resolution itself is fully automatic.
ALIAS map handles only ambiguous short forms (e.g. MPS → Monolithic Power Systems) so the yfinance search returns the right company.
STOPWORDS filter strips NVIDIA's own products, calendar/structural words, and common tech jargon (CUDA, TensorRT, Blackwell, GTC, etc.).
Negative cache: words that resolve to nothing (or non-US) are also cached, so we never re-look up junk.
scripts/nvdev_firehose.py — the enginescripts/nvdev_state.json — seen-post ids (5k cap, dedup)scripts/nvdev_ticker_cache.json — name → {ticker,exchange,name,source}The runtime repo provides a GitHub Actions workflow that runs the engine on a */30 * * * * cron with TELEGRAM_BOT_TOKEN/TELEGRAM_CHAT_ID as secrets. First run silently seeds the state (no Telegram avalanche).
| Env var | Default | What it does |
|---|---|---|
MAX_ALERTS | 8 | Cap Telegram alerts per run |
MAX_LOOKUPS | 40 | Cap yfinance.Search calls per article (cache absorbs rest next run) |
TEST_MODE | (unset) | If 1, log to stderr; no Telegram, no state/cache write |
NVDEV_USER_AGENT | (generic) | Override the polite UA string |
The cache JSON is human-editable. To override a wrong resolution or add an alias by hand:
{
"some company": {"ticker": "ABCD", "exchange": "NMS", "name": "Some Company Inc.", "source": "manual"},
"wrong match name": null
}null = negative-cached, will be skipped on future runs.
🟢🟪 NVIDIA Developer — new post
NVIDIA Transitions to 800V HVDC for 1MW AI Racks
作者: Mathias Blake et al
发布: 2025-05-20
🎯 组合内标的(高信号):
• Infineon → IFNNY
• MPS → MPWR
• Navitas → NVTS
• STMicroelectronics → STM
• Texas Instruments → TXN
• Eaton → ETN
• Schneider Electric → SBGSF
• Vertiv → VRT
• TSMC → TSM
• Coherent → COHR
• Lumentum → LITE
• Marvell → MRVL
🔍 其他识别到的可交易标的:
• ROHM → ROHCY (ROHM Co., Ltd.)
• Flex Power → FLXP (Flex-Power Inc.)
https://developer.nvidia.com/blog/...strategic-partner-firehose — SEC 8-K / 13D partner detection (downstream)pr-wire-firehose — vendor newsroom RSS (parallel signal source)insider-firehose — Form 4 buys (corroborating signal)~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.