Swift VSX Support, Biome Type Inference, Agent Guardrails — Dev Signal
Swift VSX Support, Biome Type Inference, Agent Guardrails
Share:
Tool of the Week
Swift extension lands on Open VSX Registry
Swift extension now in Open VSX Registry enables automatic installation across Cursor, VSCodium, and other LSP-compatible editors without manual setup.
Removes friction for cross-IDE Swift development; agentic IDEs can auto-install the extension, and you get code completion, debugging, and test explorer in any Open VSX-compatible editor. Workflow expands beyond Xcode without extra toolchain management.
Replaces manual extension installation and editor-specific configuration. Requires only an Open VSX-compatible editor (Cursor, VSCodium, AWS Kiro, Google Antigravity all supported). Worth adopting now if you're already in one of these editors; no blocking concerns.
“The Swift extension for VS Code is now officially available on the Open VSX Registry, the vendor-neutral, open source extension registry hosted by the Eclipse Foundation.”
“agentic IDEs like Cursor and Antigravity to automatically install Swift, with no manual download required”
“The extension adds first-class language support for projects built with Swift Package Manager, enabling seamless cross-platform development on macOS, Linux, and Windows”
swiftide-toolinglspopen-vsxcross-platform
Dev Signal
Get issues like this in your inbox — free, every weekday.
Quick Signals
Biome v2 adds type inference without TypeScript
Type-aware linting rules now work without the TypeScript compiler installed, using Biome's own type inference engine with 75% detection parity on floating promises at lower performance cost.
Eliminates TypeScript as a hard dependency for type-checking lints, reducing install size and CI overhead. Multi-file analysis unlocks rules that require cross-module context, previously impossible in v1.
Replaces typescript-eslint for type-aware rules if detection rates reach production confidence; requires migration via `biome migrate --write` command and review of breaking config changes. Opt-in file scanner means existing v1 setups won't slow down. Worth adopting now for formatting/linting speed gains; type-inference rules still maturing (75% recall is preliminary).
“the first JavaScript and TypeScript linter that provides type-aware linting rules that doesn't rely on the TypeScript compiler”
“our noFloatingPromises rule, which is based on our new type inference work, can detect floating promises in about 75% of the cases that would be detected by using typescript-eslint, at a fraction of the performance impact”
“It's opt-in; which means migrating from v1 to v2 won't significantly affect the performance of formatting and linting your projects”
“A full scan (which scans all your project files and node_modules) is performed only when project rules are enabled”
lintingtype-inferencetypescriptbiomemigration
Durable Object Facets load agent code with storage
Run AI-generated Durable Object classes inside a supervisor isolate, each with isolated SQLite storage and request interception for logging, metering, and control.
Enjoying Dev Signal? Get every issue in your inbox.
Free forever · 3 issues a week · One-click unsubscribe
Eliminates the choice between disposable agent code (no persistence) and pre-provisioned infrastructure (no containment). Developers can now sandbox persistent user-generated applications with fine-grained storage limits and observability baked in.
Replaces the pattern of forwarding requests from a Durable Object to external storage APIs. Requires Cloudflare Workers environment; available in open beta now. Worth trying immediately if you're building any form of code generation → persistent app platform. Syntax is straightforward: `this.ctx.facets.get()` with dynamic class loading.
“Durable Object Facets allow you to load and instantiate a Durable Object class dynamically, while providing it with a SQLite database to use for storage”
“The facet gets its own SQLite database, which it can use via the normal Durable Object storage APIs. This database is separate from the supervisor's database”
“isolates, not containers. Isolates are much lighter-weight than containers, and as such, can load 100x faster using 1/10 the memory”
Lightweight TF-IDF detectors outperform LLM judges by 4–8x at catching agents that falsely claim task completion, with 3,300x lower latency.
Agent silent failures—tasks reported done but actually incomplete—corrupt production monitoring. Relying on LLM judges to catch this costs you latency and misses failures; domain-calibrated statistical detectors are the practical alternative.
Replace LLM-based agent completion verification with task-specific TF-IDF triage. Requires baseline labeling on your domain (tau2-bench: AUROC 0.83; AppWorld: 0.95 achieved). Worth deploying now as monitoring layer—no latency penalty, proven higher recall.
“False success is common but varies by setting: 45--48% of failures in single-control tau2-bench domains, 3% in dual-control telecom, and 75.8% among AppWorld self-assessing coding-agent trajectories”
“no configuration across 5 judges, 5 prompt strategies, and full task specifications exceeds AUROC 0.65 on tau2-bench”
“Lightweight TF-IDF detectors achieve task-disjoint AUROC 0.83 on tau2-bench and 0.95 on AppWorld, recovering 4--8x more false successes than the best judge at the same flag rate with 3,300x lower latency”
Community trains reasoning models on free Kaggle TPUs
Google's Tunix hackathon produced reproducible recipes for adding chain-of-thought reasoning to small models (1B–2B params) using SFT + preference optimization + GRPO, all runnable in 9 hours on constrained hardware.
Developers can now train reasoning capabilities into small models without massive compute budgets. This shifts reasoning training from black-box frontier models to DIY post-training workflows with published techniques, enabling domain-specific reasoning (medical, legal, chemistry, robotics) on accessible infrastructure.
Replaces waiting for frontier model reasoning with self-service post-training on Gemma 2B/3 1B. Requires Tunix library (open-source), Kaggle TPU access (free), curated reasoning datasets (~33k–70k samples), and custom reward functions (LLM-as-judge or TF-IDF). Winner techniques are battle-tested; ready to try now with published code and Colab tutorials.
“over 11,000 entrants and 300+ high-quality submissions proved that decent reasoning training can be done by the community even with a very limited compute budget”
“Kaggle TPU v5e-8 for 9 hours”
“trains Gemma models to produce structured reasoning by combining Supervised Fine-Tuning (SFT) with GRPO, driven by a novel rubric-based LLM-as-judge reward system”
“Tunix and free Kaggle TPUs, developers can now achieve strong results on accessible hardware”
Choose where data lives—global, multi-region, dual-region, or single-region—with predictable pricing ($0.025/GB/month for multi-region) and zero egress fees.
Developers can now satisfy GDPR and data sovereignty constraints without building custom replication logic. Multi-region buckets replace hand-wired dual-region setups and cost less once you need two or more regions in a geography.
Ready now. Use `tigris mk my-bucket --locations eur` or the dashboard Advanced Settings to configure at creation time. Replaces manual S3 cross-region replication patterns. Requires updating bucket creation calls; existing buckets can migrate via dashboard Settings panel.
“Multi-region is $0.025/GB/month with zero egress fees and no per-geo premiums, so your bill stays predictable even as you add regions.”
“Consistency is tied directly to bucket location type, so you always know how reads behave in and across regions.”
“Multi-region buckets keep copies of your data across a group of regions within a chosen geography (for example usa or eur).”
“at $0.025/GB/month flat, it becomes cheaper than dual-region once you need two or more regions in a geography.”
“Use the new tigris mk commands and updated dashboard flows to create and manage location-aware buckets.”