
Hacker
N DIVIJ .
@div45
Follower of Jesus Chirst | AI Agentic Builder | 2+ years exp
Elsewhere
Stack
- TypeScript
- Python
- Rust
- AI Agents
- Anthropic
- OpenAI
- LLMs
- Fine-tuning
- UI/UX
Projects
- Top 50
Clearance: which Go vulnerability alerts are real, with proof for the rest
Graph HacksOct 2026
When a new security flaw is found in a Go library, scanners raise an alarm for every service that uses that library. But using a library doesn't mean you use the broken part of it. If the flaw is in the hammer and your code only ever uses the screwdriver, you're safe. Scanners can't tell the difference, so engineers spend hours on each alarm tracing through code, and most turn out to be false. Clearance does that tracing for you. For every service and every alarm it gives a clear answer. If you are really exposed, it shows the exact line of your code that reaches the broken function, the team that owns that line, and the version to upgrade to. If you are not exposed, it proves it, and writes that proof as a standard OpenVEX file you can hand to auditors or feed back into your scanner. If it can't be sure, for example because a scan failed, it says so instead of guessing. Ran it on 20 real Grafana services. A normal scanner raised 278 alarms. Clearance proved 65 of them false, pinned the 195 real ones to the exact line and the owning team (14 upgrades fix them all), and honestly left 18 open where the scan failed. When I fed Clearance's proof file to Trivy, a scanner teams already use, its alarms dropped from 308 to 243. Under the hood, everything lives in one FalkorDB graph: the services, the library versions they use, the published flaws, the call chains govulncheck finds in each service's code, and who owns each file. The question "does my code reach the broken function?" is a path question, so every answer is a shortestPath query in Cypher. Without the graph there is no answer. Each verdict is stored in the graph too, so when code changes, the new verdict replaces the old one and the history stays. People can ask questions in plain English through a built-in AI agent, and other AI agents can use the same graph over MCP. Every query is read-only, so no agent can change anything. Teams can also run Clearance on their own code in CI: the build fails only when the vulnerable code is really reachable, and I included a working GitHub Actions example. Built with TypeScript, FalkorDB Cloud, govulncheck, OSV, OpenVEX, Trivy, MCP and OpenAI, deployed on Vercel. Built in coordination with an AI coding assistant, Claude Code.