Security engineer. Rust. AI tooling.
I make the Light Architects platform — AI agents for Claude Code that do security reviews, enforce code quality, run investigations, and keep memory across sessions.
One plugin. Six agents. Each one does a different job.
/plugin install lightarchitects@light-architects
| Agent | Job |
|---|---|
| CORSO | Catches security issues, enforces quality gates, blocks bad code |
| EVA | Runs CI/CD, manages deploys, keeps session memory |
| SOUL | Knowledge graph — your agents remember things between conversations |
| QUANTUM | Digs into bugs, pulls from multiple sources, tests hypotheses |
| SERAPH | Pentests your code within defined scope — won't go rogue |
| AYIN | Traces every call, flags anomalies, tells you what broke and when |
Pick a team composition. Swap anytime.
/SQUAD security # pentest + appsec + forensics
/SQUAD code_review # quality + logic checks
/SQUAD research # investigation + multi-source
/SQUAD solo # just quality gates + memory
12 presets. Switch mid-session.
Rust. MCP over stdio. Native binaries under 20MB each. Runs as a Claude Code plugin.
| Repo | PR | What |
|---|---|---|
| anthropics/claude-code-security-review | #76 | Deprecated model was silently killing AI filtering — fixed |
| anthropics/claude-code-security-review | #77 | Supply chain security category (CVEs, typosquatting, lock file integrity) |
| anthropics/skills | #498 | 4-phase security audit skill |
Sr. Technical Support Engineer at Palo Alto Networks (Cortex XSOAR/XSIAM) during the day.
After hours I write the platform above. 7 Rust MCP servers, a typed SDK, a gateway with routing presets, and a knowledge graph that persists across sessions. 50K+ lines, 1,400+ tests, built solo. Every agent holds itself to the same standards I'd hold a team to — no .unwrap(), no panic!(), complexity limits enforced, clippy pedantic as errors.
I built all of it because I needed it. The security scanning exists because I review code daily. The investigation agent exists because I debug production incidents. The knowledge graph exists because I got tired of re-explaining context every session. Nothing here is theoretical.
Languages used: Rust / Python / TypeScript / Lean 4