Anthropic just dropped an 11-page PDF on killing security bugs before they're found - the 7-step Claude method that makes vulnerable code unreachable: the shift: stop hunting bugs one by one. shrink what an attacker can reach and every unknown bug inside it gets harder to hit the formula: Attack Surface = Reachable - Needed here's the method in 7 steps: step 1 → classify - public, gated or private. it decides what "too reachable" means step 2 → inventory - every entry point, every option. Anthropic's run mapped 17,000+ step 3 → defaults - code that runs out of the box but nothing uses. gate it step 4 → before login - code any stranger can trigger pre-auth. 93 needless surfaces in 11 gated projects step 5 → combos - 94 option combos that silently switched security off. reject them at load step 6 → prove it - a live trace for every finding. tests pass identical or it doesn't ship step 7 → 3 passes - fresh context each pass, short handoff, a human decides the rest the result: 25 projects, 5M+ lines, 493 issues. one combo that silently dropped a certificate check had shipped for 21 years this 11-page PDF has the 3 pass prompts, a ready skill file and the report format bookmark it before your next deploy ↓

you pay $200 a month for Claude Max. $160 of it is Claude waiting for you Anthropic measured it on their own engineers: 80% of every session, the agent just sits idle. the ones who fixed it now get 6x more agent hours than they work what they do differently: → they write the rules once. Anthropic's agents run on 8,500 standing instructions → they define "done" before the run starts → they let one agent review another → they hand off whole task types: tests, upkeep, migrations all of it maps to 8 settings already in Claude Code: 01 auto mode → no more approving every command 02 /goal → Claude keeps going until the condition is met 03 CLAUDE.md → say it once, not every session 04 plan mode → approve one plan, not 15 questions 05 reviewer subagent → read a verdict, not every line 06 Notification hook → get pinged only when it needs you 07 claude --bg → run sessions in the background 08 /loop → schedule the boring stuff start with 03, 06 and 01. takes 10 minutes the prompt below sets it all up for your repo ↓

THIS IS INSANE Anthropic just dropped a goldmine of 20 Claude Code plugins. $29,077 worth of tools, FREE on GitHub paid marketplaces sell a skill for $19 and an agent for $39. these repos ship 970 skills and 273 agents for $0 1.8M stars combined, picked from 68,900 repos tagged claude-code most people still run Claude Code bare: no memory, no team, no taste, burning tokens on raw logs 20 GitHub projects you should open first: 1. Everything Claude Code → 71 agents, 311 skills and 97 commands in one harness. the closest thing to an agent OS github.com/affaan-m/everythi… 2. agent-skills → 25 production skills by Addy Osmani: spec, plan, build, verify, review, ship github.com/addyosmani/agent-… 3. Ruflo → runs agent swarms across Claude Code and Codex with shared memory between them github.com/ruvnet/ruflo 4. agents → 202 agents, 185 skills and 105 commands packed into 94 installable plugins github.com/wshobson/agents 5. claude-mem → records what your agent did, summarizes it and feeds it back next session github.com/thedotmack/claude… 6. planning-with-files → keeps the plan in markdown on disk, so it survives /clear, compaction and crashes github.com/OthmanAdi/plannin… 7. Graphify → turns code, docs and SQL schemas into a knowledge graph. parsed locally, nothing leaves your machine github.com/Graphify-Labs/gra… 8. Understand-Anything → an interactive map of your codebase with search, guided tours and diff impact analysis github.com/Egonex-AI/Underst… 9. ponytail → makes Claude code like the laziest senior dev. 53% less code and 45% fewer tokens in its own Opus 5.5 benchmark github.com/DietrichGebert/po… 10. rtk → a Rust CLI proxy that cuts 60-90% of the tokens from common dev command output github.com/rtk-ai/rtk 11. headroom → compresses logs, tool outputs and JSON before they reach the model github.com/headroomlabs-ai/h… 12. context-mode → sandboxes tool output. 315 KB in, 5.4 KB to the model github.com/mksglu/context-mo… 13. UI UX Pro Max → 79 styles, 192 palettes and 74 font pairings, so your UI stops looking AI-made github.com/nextlevelbuilder/… 14. taste-skill → gives Claude art direction, real reference photos and a browser check at several screen widths github.com/Leonxlnx/taste-sk… 15. video-shotcraft → turns your real pages into a cinematic product video in Remotion, synced to the beat github.com/Vincentwei1021/vi… 16. Agent-Reach → lets your agent read YouTube, GitHub, Reddit and the web from one CLI, zero API fees github.com/Panniantong/Agent… 17. cc-switch → a desktop app to switch providers in one click and manage MCP, skills and prompts in one place github.com/farion1231/cc-swi… 18. pm-skills → 69 product skills and 42 chained workflows, from discovery to launch github.com/phuryn/pm-skills 19. claude-skills → 380+ skills across engineering, marketing, finance, compliance and C-level advice github.com/alirezarezvani/cl… 20. career-ops → scans job boards, scores every role against your CV and drafts the application. you still press Submit github.com/career-ops-hq/car… start with 3, not 20: one for memory, one for tokens, one for your actual job run the same task before and after. keep only what makes the result better or cheaper every plugin you add loads into context, so more isn't free save this before you spend another weekend building something that already exists and repost it for the friend who's still paying $19 a skill ↓

Claude Opus 5.5 just reverse-engineered my PS5... 1 prompt + 0 lines of code from me = a PS5 that takes voice orders: wakes up, downloads games, plays any video on demand full guide, step by step: step 1 → put your Mac and PS5 on the same Wi-Fi step 2 → paste the prompt below into Claude Code. it finds the console by itself step 3 → it digs into how the PS5 talks over the network and writes its own controller in code step 4 → approve one setup screen on the console. last time you touch the controller step 5 → just talk. it watches the screen and clicks through menus like a human the result: your PS5 now has a terminal, and Claude holds the keys... p.s. no jailbreak, no mods, nothing installed on the console. all undoable from the PS5 menu try it tonight, prompt below ↓

Anthropic just made your prompts obsolete one SKILL.md now runs in Claude, ChatGPT, Cursor and Grok Bot these 5 skills make your agent feel unfair: → caveman (114K★) - cuts token bloat without losing the fix github.com/JuliusBrussee/cav… → superpowers (300K★) - plan first, test first, prove it works github.com/obra/superpowers → supply-chain-risk-auditor - flags risky dependencies before they enter your build github.com/trailofbits/skill… → agentic-actions-auditor - finds PR-to-agent prompt injection paths in your CI github.com/trailofbits/skill… → skill-creator (180K★) - builds, tests and tunes skills that actually trigger github.com/anthropics/skills stop collecting prompts. start building skills the poster shows how to write your own SKILL.md install the skills tonight. then see how an xAI engineer scales to 200+ agents in the article below ↓

Anthropic just dropped a 15-page PDF on building your own Claude Duet - 8 layers to an agent that works while you sleep and texts you first: the formula: Proactive Agent = Context + Watch + Judgment + Gate here's how to build your own in 8 layers: layer 1 → context - one thread + your rules stored word for word. no memory without a source layer 2 → watch - one event schema for every app. a dedupe key turns 14 error pings into 1 event layer 3 → judge - anomaly score in code, then the model fills an 8-field rubric. no evidence link, no P0 layer 4 → speak - 1 brief at 8am. alerts only for P0. max 3 interruptions a day layer 5 → prepare - every overnight task gets a sandbox, a turn budget and a finish line layer 6 → gate - every action is 1 packet: exact payload, what can't be undone, expiry. nothing runs without your yes layer 7 → defend - fetched text is data, never orders. a broker holds the passwords, not the model layer 8 → learn - every skip raises the threshold 15%. P0 has a floor it can never learn past the result: replay 30 days of real events before it texts anyone. miss 1 P0 and it doesn't ship this 15-page PDF has the schemas, the pseudocode and a 10-day build path bookmark the PDF to build your own - the full course on using Duet is in the article below ↓

Claude Duet now runs your startup while you sleep Anthropic's new agent doesn't wait for a prompt. it texts you first here's the prompt I give it: → GitHub, errors, payments and support watched overnight → bug fixes and replies to angry customers drafted by morning → nothing ships until you tap approve your cofounder sleeps. this one doesn't paste this promt before bed. wake up to a brief, not a fire ↓

REA + Opus 5.5 is f*cking dangerous... a 512 MB app to change your keyboard lights. replaced by 15 lines of Python now you just tell your AI agent what you want, and your keyboard listens no app. no settings menu. no "update required" type "evening mode: warm and dim" and the whole desk goes amber ~30 seconds. watch one chat message change the lights ↓