Project ideas from Hacker News discussions.

Claude Code CLI was broken

📝 Discussion Summary (Click to expand)

1. Anthropic’s Tool Was Broken by a Basic Changelog Format Change

Users were stunned that Claude Code’s entire CLI crashed because its changelog changed a version line from ## 2.1.0 to ## 2.1.0 (2026-01-07), revealing a complete lack of parsing error handling and testing. - hughes: “Genuinely curious how a date in the subheader of a changelog could have broken the CLI... it seems changelog.md is assumed to be structured data and parsed at startup, and there are no tests to enforce the changelog structure” - mvdtnz: “The nature of the bug (a change in format for a changelog markdown file causes the entire app to break) and the testing it would have taken to uncover it (literally any) makes this one especially embarrassing for Anthropic” - Y_Y: “Ah yes, markdown, the ultimate structure for machine-readable data”

2. Claude Code Itself is Allegedly “Vibe-Coded” with LLMs

Many commenters connected the bug to the reported practice of the Claude Code team using LLMs to write most of the tool itself, calling into question its software quality standards. - frays: “Claude Code creator said Claude wrote 100% of his code last month” - cozzyd: “This sounds exactly like the type of thing you would expect an LLM to do” - frays: “I landed 259 PRs -- 497 commits, 40k lines added, 38k lines removed... The author has written an evangelical book about vibe coding” - lukev: “Lines of code never correlated with quality or even progress. Now they do even less”

3. AI Coding Tools Enable Unsafe Development & Security Risks

A parallel debate erupted about the safety and sustainability of “vibe coding,” with warnings that ignoring code understanding and review is leading to vulnerable systems. - SamInTheShell: “It is only funny until that vibe coder is building the data warehouse that holds your data and doesn’t catch the vulnerability that leads to your data leaking” - jennyholzer4: “Why would I patronize a business that hires vibe coders? I would hope their business fails if they have such pitiful security” - Bridged7756: “I fail to see how more written code by some dumbass AI that gets things wrong half the time is going to make the job less draining to me”

4. Competitor Tools Like Opencode Are Gaining Interest

Amidst the frustration with Claude Code’s bugs, users discussed alternatives like Opencode, noting its openness and ability to use existing subscriptions. - wiseowise: “It is open source, to start with” - viraptor: “It's about the same as CC. You can use subscriptions and API. It works well with basically all the providers as well - no need for hacks over Claude-like endpoints” - convenwis: “This is interesting because Anthropic seems to allow Opencode to do this but no one else. And the lead on opencode won’t comment”


🚀 Project Ideas

Jailbreak-Resistant Agent Sandbox

Summary

  • A "thin jail" or specialized container environment designed specifically for LLM agents like Claude Code.
  • Orchestrates persistent nullfs or bind mounts to specific project directories while strictly enforcing network and filesystem blacklists at the OS level.

Details

Key Value
Target Audience Security-conscious developers using AI agents
Core Feature Automated container/jail provisioning with fine-grained resource limits
Tech Stack Docker/Podman, FreeBSD Jails, or LXC
Difficulty Medium
Monetization Hobby

Notes

  • Addresses user concerns about agents running rm or searching the entire $HOME directory: "This is why I run claude inside a thin jail... It then wanted to search my entire home directory."
  • Provides the "deterministic, procedural component" for permissions that users felt the LLM was bypassing.

OpenCode Subscription Bridge

Summary

  • A proxy service or local utility that allows users to use their existing "Pro" subscriptions (Claude AI, ChatGPT Plus) with open-source coding agents (OpenCode, Qwen).
  • It handles the authentication handshakes ("opencode auth login") and system prompt injection required to make third-party models compatible with official tools.

Details

Key Value
Target Audience Developers who dislike official CLIs but pay for subscriptions
Core Feature Headless auth and protocol translation for subscription-to-API usage
Tech Stack Python, Puppeteer/Playwright (for auth session persistence)
Difficulty Medium
Monetization Revenue-ready: One-time license per user

Notes

  • Solves the frustration of hitting API limits or being forced into a specific CLI: "Hopefully it'll allow me to switch between CC and Codex easily too."
  • Leverages the "You are Claude Code" system message trick discovered by the HN community to unlock compatibility.

Agent-Watch: Output Quality Auditor

Summary

  • A dashboard for managers and senior devs to monitor the "churn" and quality of code produced by 10x/100x agentic workflows.
  • It visualizes "Line Churn vs. Progress" to identify if an agent is simply "reverting corrections" or "switching between implementations" across hundreds of commits.

Details

Key Value
Target Audience Engineering Managers, CTOs
Core Feature commit-by-commit similarity analysis and "progress vs. churn" metric
Tech Stack Python, GitPython, React
Difficulty Medium
Monetization Revenue-ready: SaaS subscription per repo

Notes

  • Responds to the skepticism around massive PR counts: "260 PRs a month means the cycle... is happening once per hour... are the wheels on the road?"
  • Directly measures the "vibe-loss" and "churn" frustrations expressed by users like mktemp-d and lelanthran.

Read Later