IdeaWell

Project ideas from Hacker News discussions.

Inside ZCode: Silently uploading your Git history to the cloud

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Privacy and data‑exfiltration concerns
Many commenters worry that AI agents silently upload code, secrets, or full repository histories without clear consent.
- “Glad my instinct to isolate it helped me, but I feel sorry for anyone whose secrets, etc. got vacuumed up by Ziphu.” – evanjrowley
- “Super sketchy though if it is in fact silently uploading full git history of every user's projects.” – loh
- “The thing that streams my code into the cloud… uploads my code into the cloud?! I didn't sign up for this!” – ThouYS

2. Distrust of closed‑source harnesses; preference for open‑source alternatives
Trust is tied to source availability; closed tools are seen as risky, while open‑source harnesses (even with flaws) are favored.
- “People have found many nasties embedded in Claude Code… You can't trust a closed source harness. You can barely trust an open source one.” – nullbio
- “Closed source agents are a red flag no matter if its China or America. Always use an open harness with a good reputation…” – tancop
- “I wouldn't list Opencode as 'good reputation'… they seem to lack the engineering culture necessary for a 'good reputation' stamp.” – hypfer

3. Need for sandboxing, isolation, and learning from past incidents (e.g., Grok Code saga)
Users advocate running agents in sandboxes/containers and treating the episode as a lesson to avoid repeating mistakes.
- “You could always sandbox it or run it in a container.” – DaSHacka
- “It is dumb to run any proprietary software without a sandbox, especially LLM‑powered.” – codedokode
- “They learned nothing from the Grok Code saga. If anything, that should have been a learning lesson to NOT trust harnesses, especially new ones.” – denysvitali

🚀 Project Ideas

OpenCode Guard

Summary

  • An open-source AI code agent harness that runs locally with strict file‑system permission scopes and optional self‑hosted LLM backends, preventing covert data uploads.
  • Core value: full transparency and control over what the agent can read, write, and send over the network.

Details

Key Value
Target Audience Developers and teams concerned about AI‑agent data leakage
Core Feature Permission‑based file access sandbox + pluggable LLM providers (self‑hosted or API)
Tech Stack Rust (core), Tauri for optional GUI, WASM plugins for model adapters, Config via TOML
Difficulty Medium
Monetization Hobby
#### Notes
- HN commenters repeatedly asked for an open harness that doesn’t silently upload repos (e.g., “We need not only open weight models, but open source harnesses as well” – loh).
- Provides a concrete way to regain trust after incidents like ZCode’s repo upload scandal, sparking discussion on safe AI‑assisted coding.

AgentSandbox

Summary

  • A lightweight container‑based sandbox that launches any AI code harness inside a confined filesystem and network namespace, logging every host access attempt.
  • Core value: enforce zero‑trust execution of AI agents without modifying the harness itself.

Details

Key Value
Target Audience Power users, security‑conscious developers, CI pipelines
Core Feature Filesystem whitelist/blacklist, network egress control, real‑time audit log UI
Tech Stack Go (container runtime), eBPF for syscall interception, React + Electron for dashboard
Difficulty High
Monetization Revenue-ready: SaaS tiered (free limited logs, $9/mo for unlimited retention)
#### Notes
- Commenters suggested sandboxing agents (“You could always sandbox it or run it in a container” – DaSHacka) and expressed frustration with hidden data grabs.
- AgentSandbox gives a practical, harness‑agnostic solution that can be adopted immediately, encouraging discussion on best practices for AI agent isolation.

HarnessWatch

Summary

  • A self‑hostable observability plugin that hooks into any AI harness (via PTrace/eBPF or language‑specific SDK) to capture file reads, writes, and outbound requests, alerting on anomalous patterns.
  • Core value: continuous visibility and forensic proof of whether an agent is exfiltrating data.

Details

Key Value
Target Audience Teams using third‑party AI harnesses who need audit trails
Core Feature Real‑time telemetry dashboard, alerting on unexpected uploads, tamper‑proof logs
Tech Stack Python agent (eBPF via bcc), FastAPI backend, Vue.js frontend, optional Loki/Prometheus storage
Difficulty Medium
Monetization Hobby
#### Notes
- Many users demanded proof (“Is there actually any proof of this?” – nullbio) and wanted transparency; HarnessWatch supplies verifiable logs.
- By enabling independent verification, it addresses the trust gap highlighted in the thread and can be integrated with existing harnesses like OpenCode or ZCode.

Read Later