The interesting part of a new developer tool is not the launch announcement. It is whether you can put it through a small, measurable experiment.
Here are five current projects and topics worth testing, with a bounded first step for each. The goal is not to declare a winner; it is to find where the tool fits your workflow.
1. Claude Code: Require evidence before accepting a fix
What it is: Claude Code is Anthropic’s terminal-based coding agent. It can inspect a repository, modify files, run development commands, and respond to failures.
Why now: The Claude Code team’s recent verification write-up reached 256 points and 165 comments on Hacker News at the time of this report. The discussion reflects a practical concern: an agent saying “fixed” is not the same as reproducing the bug, running the correct tests, and checking for regressions.
Tutorial angle: Start with a small, reproducible bug in a sample repository. Write or seed a regression test that fails before the fix. Then ask Claude Code to:
- Reproduce the failure and show the command output.
- Add or update a regression test.
- Implement the smallest possible fix.
- Run both the targeted test and the full test suite.
- Review its own diff for unrelated changes.
A useful prompt is:
Fix this bug. First add a regression test and demonstrate that it fails for the reported reason. Then make the smallest change that passes it. Run the targeted test and the full suite, show both results, and review the final diff. Do not claim the issue is fixed unless the required commands exit successfully. Keep the initial repository disposable. You are evaluating the verification loop, not trying to complete a large migration on day one.
Difficulty: Intermediate
Start today:
- Install Claude Code with
curl -fsSL https://claude.ai/install.sh | bash
2. rust-anything: Port behavior before you port code
What it is: rust-anything is an assistant for incrementally rewriting an existing application in Rust. Its approach focuses on comparing the rewritten implementation with the behavior of the original.
Why now: The rust-anything repository reportedly collected 1,292 GitHub stars between its November 16 creation and this report. The interest is understandable: big-bang rewrites often hide behavioral differences until much later.
Tutorial angle: Port one small surface area, such as a single CLI command or HTTP endpoint:
- Record golden outputs from the existing implementation.
- Generate the Rust equivalent.
- Run both versions against the same inputs.
- Compare stdout, stderr, exit codes, and response bodies.
- Investigate every mismatch before accepting the port.
- Commit each port separately so it can be reverted.
Choose behavior with clear edges. Date formatting, Unicode handling, and partial failures are more useful first tests than a complicated business workflow.
Also decide what “equivalent” means before generating code. Matching a response byte-for-byte may be too strict for one endpoint and too loose for another.
Difficulty: Intermediate
Start today:
- Clone the project with
git clone https://github.com/erkamyaman/rust-anything.git
3. FrontierHarness: Evaluate the harness, not only the model
What it is: FrontierHarness is an evaluation harness for comparing models inside different agent harnesses. A harness includes the system prompts, tool loops, context management, and scaffolding surrounding a model.
Why now: FrontierHarness Eval v1.1 reached 329 points and 161 comments on Hacker News at the time of this report. That points to a gap familiar to anyone running coding agents: a model can perform well on a generic benchmark but behave differently when it has your tools, prompts, and repository.
Tutorial angle: Build a small benchmark from three representative repository tasks, such as:
- Fix a bug and add a regression test.
- Add an endpoint to an existing service.
- Refactor a module without changing its public API.
Run the same task snapshots through several model-and-harness pairings. Record:
- Task completion rate
- Test pass rate
- Number of tool calls and turns
- Time and token usage
- Unrelated changes in the final diff
- Whether the agent recovered after a failed command
Use those results to create a routing rule—for example, “use model A for test-driven fixes and model B for bounded refactors”—rather than selecting one model for everything.
Difficulty: Intermediate
Start today:
- Open the FrontierHarness Eval v1.1 quickstart and locate its first example task
4. Java4Shell: Triage OpenJDK Base64 exposure
What it is: Java4Shell is a newly disclosed OpenJDK vulnerability involving attacker-controlled Base64 data that can lead to remote code execution.
Why now: The Java4Shell disclosure reached 450 points on Hacker News, making runtime verification an immediate priority for teams operating Java services.
Tutorial angle: Treat this as an inventory and patching exercise, not an exploit exercise:
- Record the Java runtime version used by each service.
- Find the boundaries where untrusted data enters Base64 decoders.
- Compare those runtimes with the fixed versions in the vendor advisory.
- Upgrade affected services and redeploy every instance.
- Add a harmless regression test that confirms malformed or unexpected encoded input is rejected safely.
- Verify the updated runtime in the deployed environment, not only on a build server.
Inventorying decoding boundaries is more useful than searching for one suspicious function. The dangerous path may cross a controller, queue consumer, plugin loader, or configuration service.
Test only in an isolated environment. Do not run third-party proof-of-concept payloads against production systems.
Difficulty: Intermediate
Start today:
- Run
java -version 2>&1and compare the runtime version with the advisory
5. open-orange: Test Cloudflare Worker portability
What it is: open-orange is an open-source compatibility layer designed to run Cloudflare Workers and Durable Objects on AWS, GCP, Azure, or self-hosted infrastructure.
Why now: The open-orange repository reportedly reached 659 GitHub stars within days of its November 12 creation. Portable serverless runtimes are appealing because they can reduce infrastructure coupling without forcing an immediate application rewrite.
Tutorial angle: Build a deliberately small Worker that uses a Durable Object, then:
- Run it locally through
open-orange. - Exercise the Worker’s routing behavior.
- Write and read Durable Object state.
- Switch the target backend without changing application code.
- Compare responses, persistence, concurrency behavior, and error handling.
- Record any Cloudflare-specific APIs that do not translate cleanly.
Include failure cases in the comparison. Portability is more convincing when requests, timeouts, storage failures, and concurrent updates behave consistently.
Difficulty: Intermediate
Start today:
- Clone the runtime with
git clone https://github.com/rivet-dev/open-orange.git
Building in public: my lead-gen engine
My lead-gen engine discovers and enriches Hong Kong SME leads before preparing them for outreach. In the latest run, it discovered 1,776 leads, enriched 201, and contacted 231. The gap between discovery and enrichment is the clearest constraint, so the next iteration should measure successful enrichment and reply quality rather than focusing only on discovery volume.
- Discovered leads: 1,776
- Enriched leads: 201
- Contacted leads: 231
Start with this today
Pick Java4Shell because runtime exposure is quick to check.
Run:
java -version 2>&1 Compare the output with the Java4Shell disclosure and vendor guidance. If the version or deployment configuration is uncertain, check the runtime inside the actual container or deployment image rather than relying only on your local machine.
Picks of the week
- Claude Code for teams that need agents to show verification evidence.
- rust-anything for incremental, behavior-preserving Rust migrations.
- FrontierHarness for choosing models based on real harness performance.
- Java4Shell for immediate OpenJDK runtime triage.
- open-orange for testing whether a Worker can move between clouds.
Which of these would you put through a five-minute test first, and what result would convince you to adopt it?
Connect
If this kind of post is useful, the easiest way to support the work is to:
- Star / follow on dev.to (you're already here 🙂)
- Follow on X: @devkingov
- Reach out for HK-based dev work — .NET / Azure / system integration / IT security: studio.resurrects.co or email devkingov@gmail.com
- Subscribe to weekly HK tech posts → studio.resurrects.co/blog (one email a week, no spam)

