[{"data":1,"prerenderedAt":49},["ShallowReactive",2],{"$fTTNp6T5g0LFkZC24PH8MUEGWSFb_K8102Fp7DnugZGI":3},{"date":4,"generated_at":5,"picks":6,"candidates_scanned":47,"candidates_scored":48},"2026-04-21","2026-04-21T06:00:00.000000+00:00",[7,21,34],{"rank":8,"title":9,"source":10,"url":11,"category":12,"tldr":13,"score":14,"scores":15,"why":20},1,"v2.1.116","Claude Code Releases","https://github.com/anthropics/claude-code/releases/tag/v2.1.116","Release","- `/resume` on large sessions (40MB+) is now up to 67% faster — if you work with long-running sessions, this is immediately noticeable\n- MCP startup is faster when you have multiple stdio servers: the `resources/templates/list` call is now deferred until you actually type `@` to reference a resource\n- Security fix: sandbox auto-allow no longer bypasses the dangerous-path safety check for `rm`/`rmdir` targeting `/`, `$HOME`, or other critical system directories\n- The thinking spinner now shows inline progress text (\"still thinking\", \"thinking more\", \"almost done thinking\") instead of a separate hint row\n- `/config` search now matches option values — searching \"vim\" finds the Editor mode setting; `/doctor` can now be opened while Claude is mid-response",69,{"direct_claude_relevance":16,"practical_utility":17,"novelty":18,"source_credibility":19},25,20,11,13,"A substantive patch release with 20+ changes spanning performance, UX, security, and bug fixes. The 67% /resume speed improvement on large sessions is the headline, but the security fix for sandbox auto-allow bypassing dangerous-path checks is the most important change for anyone using Claude Code in agentic workflows. The Ctrl+Z hang fix when launched via npx/bun and the Kitty keyboard protocol fixes round out a release that touches nearly every part of the tool.",{"rank":22,"title":23,"source":24,"url":25,"category":12,"tldr":26,"score":27,"scores":28,"why":33},2,"[Release] anthropics/claude-agent-sdk-python: v0.1.64","GitHub anthropics/claude-agent-sdk-python","https://github.com/anthropics/claude-agent-sdk-python/releases/tag/v0.1.64","- The Python SDK now has full `SessionStore` support at parity with the TypeScript SDK — store, load, fork, and list sessions against any backend you implement\n- Three ready-to-copy reference adapters included: S3 (JSONL part files), Redis (RPUSH/LRANGE), and Postgres (asyncpg + jsonb) — copy whichever file you need into your project, no extra wheel dependency\n- A 13-contract conformance test harness at `claude_agent_sdk.testing.run_session_store_conformance` lets third-party adapter authors verify their implementation is correct\n- Install: `pip install claude-agent-sdk==0.1.64`",67,{"direct_claude_relevance":29,"practical_utility":30,"novelty":31,"source_credibility":32},27,21,9,10,"Python developers building agent pipelines finally get the session persistence story that TypeScript developers got in v0.2.113 (covered April 18). The three reference adapters — S3, Redis, and Postgres — are copy-in templates rather than installed dependencies, which is the right design: you own the code, you control the schema. The conformance test harness is particularly useful for teams building their own backends against a database or queue system not covered by the reference implementations. Novelty is penalized because the SessionStore concept itself was covered in the TypeScript SDK three days ago, but the Python implementation is fully independent and the audience is distinct.",{"rank":35,"title":36,"source":37,"url":38,"category":39,"tldr":40,"score":41,"scores":42,"why":46},3,"Claude Prompt Caching in 2026: The 5-Minute TTL Change That's Costing You Money","Dev.to Claude","https://dev.to/whoffagents/claude-prompt-caching-in-2026-the-5-minute-ttl-change-thats-costing-you-money-4363","Guide","- Anthropic dropped prompt cache TTL from 60 minutes to 5 minutes in early 2026 — a batch worker that previously got 20 reads per cache write now gets ~2, cutting effective savings from 84% to ~52%\n- Who gets hit hardest: batch pipelines with gaps longer than 5 minutes, cron agents on 15-30 minute schedules, and chat apps where user sessions go idle",57,{"direct_claude_relevance":43,"practical_utility":17,"novelty":44,"source_credibility":45},24,8,5,"The TTL reduction from 60 to 5 minutes is a real change that meaningfully alters the economics of Claude API workloads — particularly for batch pipelines and scheduled agents that were designed around the old TTL. The article works through the before/after math concretely with token counts and dollar figures, and identifies the workload patterns most affected. The architecture patterns for working around it (warming within 5-minute windows, restructuring batch schedules) are actionable today. Dev.to source limits the credibility score, but the cost analysis is independently verifiable.",26,12,1776747801040]