[{"data":1,"prerenderedAt":49},["ShallowReactive",2],{"$ff691TRPdTVBpbZS_PFB1UNlG4DUEXTC3PZVuD2Xl5vs":3},{"date":4,"generated_at":5,"picks":6,"candidates_scanned":47,"candidates_scored":48},"2026-03-27","2026-03-27T05:30:00.000000+00:00",[7,21,35],{"rank":8,"title":9,"source":10,"url":11,"category":12,"tldr":13,"score":14,"scores":15,"why":20},1,"v2.1.85","Claude Code Releases","https://github.com/anthropics/claude-code/releases/tag/v2.1.85","Release","- Hooks just got a lot smarter: you can now add an `if` field using permission rule syntax (e.g., `Bash(git *)`) so hooks only fire on the tool calls you actually care about — no more spawning a process for every single command\n- PreToolUse hooks can now answer an `AskUserQuestion` by returning `updatedInput` alongside `permissionDecision: \"allow\"` — this is the missing piece for headless CI integrations that need to inject answers without a human at the keyboard\n- The `/compact` command no longer crashes when the conversation is so large that the compact request itself exceeds the context limit — it quietly handles the overflow\n- Timestamp markers now appear in transcripts whenever `/loop` or `CronCreate` tasks fire — finally you can see exactly when a scheduled task ran in a long session\n- Deep link queries (`claude-cli://open?q=…`) now support up to 5,000 characters, with a scroll-to-review warning for anything long\n- Plugins blocked by organization policy are now hidden from the marketplace entirely, not just disabled\n- Fixed `deniedMcpServers` not actually blocking claude.ai MCP servers — the setting now works as documented\n- Fixed `switch_display` returning \"not available in this session\" on multi-monitor setups in computer-use\n- Fixed a memory leak in remote sessions when a streaming response is interrupted\n- Fixed ECONNRESET errors during edge connection churn by using a fresh TCP connection on retry",92,{"direct_claude_relevance":16,"practical_utility":17,"novelty":18,"source_credibility":19},34,27,17,14,"This release ships two genuinely new capabilities that change how hooks and headless integrations work. Conditional hooks with the `if` field are huge for anyone running multiple hooks — you can now target exactly which tool calls trigger your hook, cutting unnecessary process spawns dramatically. The PreToolUse → AskUserQuestion bridge is the other standout: it unlocks fully headless Claude Code setups where an external system handles the UI and passes answers back via hooks. The /compact crash fix and deniedMcpServers fix both address bugs that were actively hurting production users in longer sessions.",{"rank":22,"title":23,"source":24,"url":25,"category":26,"tldr":27,"score":28,"scores":29,"why":34},2,"Exclusive: Anthropic acknowledges testing new AI model representing 'step change' in capabilities, after accidental data leak reveals its existence","Fortune","https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities/","Announcement","- Anthropic accidentally leaked the existence of a new model called \"Mythos\" through a data leak before any official announcement\n- After the leak, Anthropic confirmed they are actively testing Mythos and described it as a \"step change\" in capabilities — not an incremental update\n- No public release date or pricing yet — this is purely a confirmation that it exists and is in testing",63,{"direct_claude_relevance":30,"practical_utility":31,"novelty":32,"source_credibility":33},26,5,19,13,"\"Step change\" is strong language from a company that typically undersells. The accidental leak followed by Anthropic's official acknowledgement makes this credible, not just a rumor. Mythos isn't Claude 4 or a minor point release — Anthropic is signaling a qualitatively different capability level, though what that means in practice won't be clear until they demo it publicly. Worth knowing exists even if there's nothing actionable yet.",{"rank":36,"title":37,"source":38,"url":39,"category":12,"tldr":40,"score":41,"scores":42,"why":46},3,"[Release] anthropics/claude-agent-sdk-typescript: v0.2.85","GitHub anthropics/claude-agent-sdk-typescript","https://github.com/anthropics/claude-agent-sdk-typescript/releases/tag/v0.2.85","- New `reloadPlugins()` method lets SDK-based integrations hot-reload plugins and get back refreshed commands, agents, and MCP server status without restarting\n- Fixed a bug where PreToolUse hooks returning `permissionDecision: \"ask\"` were silently ignored in SDK mode — they now correctly prompt as expected",68,{"direct_claude_relevance":43,"practical_utility":44,"novelty":45,"source_credibility":33},28,16,11,"Small release but ships at parity with Claude Code v2.1.85 — if you're building on the TypeScript SDK and relying on PreToolUse hooks with `permissionDecision: \"ask\"`, the bug fix is directly relevant (your hooks were being ignored). The `reloadPlugins()` method matters for long-running SDK processes that need to pick up plugin changes without a full restart — a real pain point in development workflows.",46,20,1776402243340]