Axel Bitblaze 🪓

@MedusaOnchain gm medusa

gooood morningggg still unbothered, moisturized, happy in my lane, focused, flourishing, popping

Some work for you to do today itself: I was going through my .claude folder yesterday and realized something that scared me.. claude code reads your .env file the moment you open a project. every API key in there like i use binance, dune, openai, stripe - gets loaded into the conversation. and the conversation goes to anthropic's servers. i had "never read .env files" in my CLAUDE.md . thought that was enough. it's not.. CLAUDE.md is just a suggestion to claude. it follows it most of the time. but on long, complex tasks it forgets. there was a github issue in april where claude leaked env contents even with the rule in place. if you trade with API keys in .env, this isn't a maybe. one ambiguous prompt and your keys are sitting on someone else's server. 3 ways your secrets leak. > claude opens the .env file directly while exploring your project. the obvious one. > claude runs your tests. one test fails and the error log shows the full API key in plain text. claude captures everything it sees in the terminal, including that. > claude searches your code with grep. the search finds a config file that has credentials. matched lines show up in chat. nobody asked for the keys but they're there. most people only block the first one. the other two are where you actually get hurt. the fix is one file. open ~/.claude/settings.json and add this: { "permissions": { "deny": [ "Read(**/.env*)", "Read(**/*.pem)", "Read(**/*.key)", "Read(**/secrets/**)", "Read(**/credentials/**)", "Read(**/.aws/**)", "Read(**/.ssh/**)", "Write(**/.env*)", "Write(**/secrets/**)" ] } } these "deny rules" tell your computer to block claude from reading those files. it's not a polite request like CLAUDE.md. claude literally cannot open them. for the runtime leak, make a separate file called .env.test with fake values. point your tests at that instead of your real .env. now if a test fails and dumps secrets, only fake secrets show up. STRIPE_SECRET_KEY=sk_test_not_real OPENAI_API_KEY=sk-test-dummy extra layer for safety. add a pre-commit hook that scans your code for common API key patterns before they ever reach github. #!/bin/bash PATTERNS=('sk-ant-' 'sk-live-' 'ghp_' 'AKIA' 'eyJ') for pattern in "${PATTERNS[@]}"; do if git diff --cached | grep -qE "$pattern"; then echo "BLOCKED: secret found" exit 1 fi done save it as .git/hooks/pre-commit and run chmod +x .git/hooks/pre-commit. 5 minutes to set up. stays on every project from now on. if you skip this, your next prompt could put a live key in an anthropic chat log and rotating keys after a leak is the kind of work nobody plans on a friday.

Spent today deleting over 30 MCPs from my Claude Code as most of them were costing me money to do nothing.. ( realised it a bit late ) each one was eating 200-400 tokens of context before i'd even typed a prompt. kept 6. the rest were bookmarks i installed once and forgot about.. already seeing my claude got faster and sharper with the outputs

this week is important for bitcoin we’re holding above the 100D EMA (72.6k) for the first time since the top but still stuck below heavy sellers at 78K - 80k 80k is the level that needs to flip imo we’re mid decision rn break above = momentum comes back reject = more chop

@nikitabier then what are they here for ? porn?

The most snoozed (i.e., muted) topics since launching the snooze feature: 1. Crypto 2. Politics 3. Iran Conflict 4. Sports 5. Business & Finance 6. Gaming 7. Artificial Intelligence 8. Videos 9. Science & Technology 10. Entertainment & Arts

i too shared this a couple days ago, now saying it again after using explicitly GPT 5.5 on Codex shorter outcome focused prompts are straight up better now the model finally cooks without the long hand holding so far the output im getting from GPT 5.5 is upto my expectations

@obsidianstudio9 been using this since last month

【速報】 Obsidian公式がAIエージェント用スキルを正式リリース😳 CEOのkepano自らが作ったobsidian-skillsが公開された。 Claude CodeにObsidianの使い方を教えるスキル集👇 ・Markdown構文の正しい扱い方 ・Bases（データベース機能）の操作 ・JSON Canvasの生成と編集 ・CLIでのvault操作 ・Webコンテンツの取り込み 5つのスキルが1セットになってる。 つまりObsidian公式が「AIエージェントにvaultを任せる」前提で設計し始めた。 この流れは確実に加速する🔥 github.com/kepano/obsidia…

Imagine running a deep research tool that beats Gemini, OpenAi and perplexity on all 3 benchmarks for $5 a month. on a small VPS that's what Onyx + Crewai + Voxtral give u its open source + self-hosted and cost effective + your data never leaves ur machine.. full article below

I automated my pump-fun launch hunting strategy with Browser Harness across 4 sites in parallel.. just claude opening tabs and reading what humans read.. like: > Pumpfun for traction signals > Dexscreener for buying pressure > X for verified mention growth > Solscan for dev reputation history scored 15 candidates across all 4. returned top 5 ranked.. works well

My Top 30 repos.. coolest + most useful + actually simplified how i work.. I have been exploring repos like crazy this week… installing one after another on claude code and codex to see what actually delivers. these 30 survived. ▫️First some claude code essentials: 1) Terminal agent itself github.com/anthropics/cla… 2) Official skills + mcp-builder github.com/anthropics/ski… 3) Master list everyone forks github.com/hesreallyhim/a… ▫️For memory + context: 4) Auto persistent memory across sessions. 46K stars github.com/thedotmack/cla… 5) live library docs before claude writes code github.com/upstash/contex… 6) long term memory layer for agents github.com/mem0ai/mem0 ▫️ Now some for research + scraping 7) reddit/x/youtube/hn scraped in one prompt github.com/mvanhorn/last3… 8) self healing browser automation github.com/gregpr07/brows… 9) reliable scraping with anti-bot built in github.com/mendableai/fir… ▫️ For trading + onchain MCPs 10) wallet, contract, token analysis github.com/crazyrabbitLTC… 11) SQL on onchain data 20+ chains github.com/kukapay/dune-a… 12) real time polymarket odds github.com/kukapay/polyma… 13) reads your live TV chart state github.com/tradesdontlie/… 14) 100+ exchanges, one server github.com/doggybee/mcp-s… ▫️Multi-agent finance frameworks 15) HKU workspace. 71 skills, 29 teams, 7 backtest engines github.com/HKUDS/Vibe-Tra… 16) 13 official binance trading skills github.com/binance/binanc… 17) automated btc 5min prediction market bot github.com/KaustubhPatang… 18) AI agents simulating buffett, munger, ackman, wood github.com/virattt/ai-hed… 19) multi-agent framework with risk manager github.com/TauricResearch… ▫️Finance infra + models: 20) open source bloomberg alternative github.com/OpenBB-finance… 21) first open source foundation model for candlesticks github.com/shiyu-coder/Kr… 22) microsoft's AI quant investment platform github.com/microsoft/qlib ▫️Skill ecosystem 23) karpathy's coding observations as CLAUDE.md. 5700 stars in a day github.com/forrestchang/a… 24) skill marketplace + mcp-builder github.com/vercel-labs/sk… 25) 1000+ community plugins indexed github.com/quemsah/awesom… ▫️Agent frameworks 26) production agent orchestration github.com/langchain-ai/l… 27) team based agents with shared task lists github.com/crewAIInc/crew… 28) microsoft's multi-agent framework github.com/microsoft/auto… ▫️For content + local 29) heygen's reels video framework github.com/heygen-com/hyp… 30) run claude style agents on your own hardware github.com/nvidia/nemoclaw make sure to bookmark this. u might need this in future…

32 Claude Code hacks broken down from beginner to pro.. good article i picked a few from this.. worth your bookmark.

Vibe Trading is now a new thing.. and learn this right now.. its open source, free, runs locally. INSTALL THIS! first, what it does is basically you describe a trade idea in plain english, it backtests it across crypto, stocks, futures, forex. takes about 2 mins. Install: pip install vibe-trading-ai basic command: vibe-trading run -p "Backtest BTC MACD strategy last 30 days" it pulls the data, runs the backtest, gives you Sharpe ratio, drawdown, win rate. no API keys needed for crypto. the bigger play. 29 pre-built agent teams. you run one and 4-5 specialist agents work together on a single task. example. crypto trading desk team: vibe-trading --swarm-run crypto_trading_desk '{"asset": "ETH-USDT", "timeframe": "1w"}' - Agent 1 analyzes funding rates - Agent 2 reads liquidation heatmaps - Agent 3 tracks stablecoin flows. risk manager reviews everything and gives you a position call. other teams worth running: - investment_committee - bull vs bear debate - quant_strategy_desk - factor research to backtest - technical_analysis_panel - 5 TA frameworks reach consensus - macro_rates_fx_desk - rates, FX, commodity macro brief ships as an MCP. plug into claude code: npx clawhub@latest install vibe-trading now your claude has 17 trading tools. you build research workflows on top. bonus. /pine command exports your backtested strategy to TradingView Pine Script. backtest here, run on tradingview. does NOT execute live trades. it's a research tool. but it's the closest open-source thing to a bloomberg terminal i've seen. repo: github.com/HKUDS/Vibe-Tra…