Application security · AI agent security · Security tooling
I research Java / Spring applications and Agent / MCP trust boundaries, and build tools that make security work reproducible. My work spans source review, vulnerability validation, coordinated disclosure, and remediation follow-up.
Website · Research notes · SoybeanSeclab
Personal and team projects, selected for practical security work and agent engineering. Stars are repository counts as of 2026-11-14.
| Project | Stars | Work |
|---|---|---|
| Phantom | 692 | Team. Browser extension for recursive asset discovery, API mapping, batch checks, custom rules, and exports. |
| AgentFlowDesk | 65 | Personal. Local Python CLI for context packs, agent instructions, run logs, diffs, checks, and review reports. |
| prtseek | 53 | Team, based on ARTEX. Four-role CTF agent with shared budgets, evidence records, interruption handling, and isolated experiments. |
| prtstrike | 48 | Team. Go command-and-control framework with a web console, peer-node coordination, and task management for authorized labs. |
| JavaSecAtlas | 2 | Team. v1.2.0 Java / Spring audit workbench with permission context, SARIF import, report generation, and AI-assisted review. |
prtseek update history · JavaSecAtlas v1.2.0 notes
- NocoDB — GHSA-hjgf-2f3h-4962: server-side request forgery via integration remote fetch. Published 2026-11-02, High / CVSS 8.5, with OracleNep credited as the reporter. The advisory describes internal-network access and integration-secret exposure, with remediation across releases 2026.08.0 and 2026.10.1.
- DeepSeek Harness — public research report #853: authentication and trust-boundary analysis of the
dsh webcontrol plane, tested on 0.1.0-rc.6. The report distinguishes local-process access from remote exposure under non-loopback configurations and proposes authentication and authorization improvements. - Private GHSA reporting — Flowise and Strapi: submitted reports through GitHub's private security advisory workflow. As of 2026-11-14, one Flowise advisory remains in draft and three Strapi reports remain in triage. These are not published advisories; technical details remain private during coordination with maintainers.
Selected identifiers from individual and team vulnerability reporting: CVE-2025-7181, CVE-2025-7175, CNVD-2026-10837, and CNVD-2025-17734.
| Contribution | Status | Change |
|---|---|---|
| VulnClaw #75 | Merged | Bounded solve-frontier recovery, fallback exploration, CLI events, and regression coverage when open intents are exhausted before the goal is complete. |
| PHP #22261 | Merged | Documentation and test-comment maintenance for ext/intl error conventions. |
| PHP #22223 | Open for review | Aligns request_parse_body() handling of max_input_vars with INI behavior, with PHPT tests and compatibility notes. |
Statuses checked on 2026-11-14.
- Application security: Java / Spring code review, authorization and business-logic boundaries, vulnerability reproduction, root-cause analysis, and remediation validation.
- AI agent security: tool execution, MCP permissions, control-plane authentication, approval flows, recovery behavior, and evidence records.
- Security engineering: Python, Java, JavaScript, Semgrep / CodeQL workflows, SARIF processing, automated reporting, and CI checks.
- Authorized security practice: incident triage, host-security checks, impact assessment, hardening recommendations, and retesting.
Technical notes, currently in Chinese:
