[{"data":1,"prerenderedAt":45},["ShallowReactive",2],{"$ftwpsBeDs0WF3MDAFnD3_5WurfdRVeweXoYRpWburNmg":3},{"date":4,"generated_at":5,"picks":6,"candidates_scanned":44,"candidates_scored":41},"2026-05-04","2026-05-04T06:00:00.000000+00:00",[7,21,32],{"rank":8,"title":9,"source":10,"url":11,"category":12,"tldr":13,"score":14,"scores":15,"why":20},1,"How I stopped Claude Code from hallucinating function names on a 4,000-file repo (with a local MCP server)","Dev.to Claude","https://dev.to/nike-17/how-i-stopped-claude-code-from-hallucinating-function-names-on-a-4000-file-repo-with-a-local-mcp-jl5","Guide","- In a 4,000-file TypeScript monorepo, Claude Code kept inventing function names that looked plausible but didn't exist — `logResponseTime`, `parseConfigFile`, `validateInput` — then calling them confidently in generated code\n- Root cause: without a real symbol graph, Claude fills gaps with statistically plausible names from training data; the larger the repo, the worse it gets\n- Fix: run a local MCP server that gives Claude Code actual symbol lookup and ranked code search, replacing grep+read with a proper index\n- Result: function name hallucinations dropped to near-zero on the same repo — author includes benchmark numbers (NDCG-style) and explicit cases where it still doesn't help",63,{"direct_claude_relevance":16,"practical_utility":17,"novelty":18,"source_credibility":19},23,22,13,5,"Claude Code hallucinating function names in large repos is a widely-reported frustration with no official fix — this post diagnoses the root cause clearly and gives a concrete local MCP server solution with benchmarks. The failure is reproducible and the fix is low-friction to try. The author also honest about where it doesn't help, which gives the advice credibility.",{"rank":22,"title":23,"source":10,"url":24,"category":12,"tldr":25,"score":26,"scores":27,"why":31},2,"How I run Claude Code 24/7 on a 2015 MacBook — the framework that survived 6 months","https://dev.to/suzu/how-i-run-claude-code-247-on-a-2015-macbook-the-framework-that-survived-6-months-296n","- A non-developer ran Claude Code on a 2015 MacBook Air for 6 months — 857 autonomous sessions, 840 self-merged PRs — and open-sourced the framework that made it survivable\n- Three problems autonomous Claude Code always hits: sessions start blank (no memory of yesterday), failures cascade silently overnight, and git becomes a graveyard of orphan branches\n- The fix is five small scripts around the existing `claude` CLI — no new agent framework: launchd scheduler, session-start/end hooks to inject/extract context, scope-guard to block protected files, and a healthcheck that opens GitHub Issues on failure\n- Open-sourced as `claude-autonomous-kit` on GitHub — each piece is under 200 lines",61,{"direct_claude_relevance":28,"practical_utility":29,"novelty":30,"source_credibility":19},24,20,12,"Autonomous Claude Code operation is something many users experiment with but few sustain — this is a field report from someone who actually ran it for six months with production data. The three failure modes identified (blank sessions, silent cascade failures, git branch graveyard) are immediately recognizable to anyone who has tried scheduled Claude Code. The framework is deliberately minimal, making it adoptable rather than aspirational.",{"rank":33,"title":34,"source":35,"url":36,"category":37,"tldr":38,"score":39,"scores":40,"why":43},3,"DeepClaude – Claude Code agent loop with DeepSeek V4 Pro, 17x cheaper","HN Claude Code","https://github.com/aattaran/deepclaude","Release","- DeepClaude plugs DeepSeek V4 Pro in as the \"thinking\" layer before Claude Code executes — planning and reasoning on the cheap model, actual code actions left to Claude\n- Claims 17x cost reduction vs. Claude-only loops; 268 HN upvotes and 110 comments with skeptical benchmarks and real comparisons in the thread",59,{"direct_claude_relevance":41,"practical_utility":29,"novelty":30,"source_credibility":42},18,9,"DeepSeek V4 Pro is newly released and pairing it with Claude Code as a cheaper reasoning preprocessor is a meaningfully different pattern from general model routing — it's a hybrid agent loop, not just task delegation. The HN thread has enough skeptical commentary and alternative comparisons to be worth reading before adopting it, which is what makes the community discussion valuable beyond the GitHub readme.",31,1777871084866]