Skip to content
View citizen204's full-sized avatar

Highlights

  • Pro

Block or report citizen204

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
citizen204/README.md

Hi, I'm Xi Chen

Cybersecurity undergraduate at Adelaide University (Bachelor of IT, Networking and Cyber Security), graduating July 2027. I work on CI/CD and software supply-chain security, mostly by building tools and then finding out where they are wrong.

Open to internships and part-time work in security, in Adelaide or remote.

Projects

ghast: a taint-analysis scanner for GitHub Actions workflows. It follows attacker-controlled data (an issue title, a comment body, a fork's branch name) across job boundaries into whatever interprets it, and scores each finding by what an attacker actually gets. A first sweep of ~1,800 workflow files, every high-severity result read by hand, exposed thirteen classes of false positive in my own tool, each now a regression test. After the fixes, 709 workflow files in 29 major repositories gave 8 critical/high findings, every one confirmed against the source. Write-up: I built a GitHub Actions security scanner, then spent longer proving it wrong.

noisefloor: a SOC triage drill that grades the evidence an analyst cited, not only their verdict. Scenarios come in pairs: the same rule fires with near-identical alert text and reaches opposite verdicts, so the exercise is reading evidence rather than recognising a format.

sentinel-toolkit: a modular defensive security toolkit. Four scanner modules feed one finding model, with JSON/HTML/SARIF reports and a Next.js dashboard. It records what each run actually assessed, and only reports a problem as resolved where a later run provably covered it.

Open source

50 merged pull requests across 15 projects, including Scapy, mitmproxy, pwntools, objection, OnionShare, dify and llama_index. Two I'd point to:

  • AutoGPT #14537, found with ghast's read-only hunt mode: a workflow handed a secret to an action resolved from a moving HEAD. Pinned it to a verified commit and set the job's token permissions to {}.
  • openlit #1324 closed an unauthenticated SSRF (CWE-918) in a production observability platform. It was a four-line deletion, made after establishing that the wildcard it removed was not needed anywhere in the app.

Before security

I built, and still support, full-stack products, and interned at Baidu Smart Cloud, where I designed account security verification flows for high-risk operations.

Python · TypeScript · SARIF / CodeQL · GitHub Actions · MITRE ATT&CK · FortiGate · Docker · AWS · PostgreSQL

LinkedIn · Email

Pinned Loading

  1. ghast ghast Public

    Taint-analysis security scanner for GitHub Actions workflows. Traces attacker-controlled data across jobs to the sink, and scores findings by what the attacker actually gets.

    Python

  2. cloud-ops-ai-agent cloud-ops-ai-agent Public

    Async Industrial Execution Engine — merging Baidu Cloud-Phone scalable ops patterns with modern AI agent design. Features bounded concurrency, 3-phase safety gateway, config-driven retry, and Trace…

    Python 1

  3. noisefloor noisefloor Public

    A SOC triage drill that grades your reasoning, not just your verdict. Paired scenarios where the same alert is a true positive and a false positive.

    Python

  4. sentinel-toolkit sentinel-toolkit Public

    A modular defensive security toolkit: pluggable scanners, unified findings, JSON/HTML reports. First module: AWS misconfiguration scanning (S3, security groups, IAM MFA).

    Python