Agents Running Agents: Codex Meets OpenCode and DeepSeek
A hello-world DeepSeek request became a safer agent-to-agent runner, a dated route preflight, and a Cost per Accepted Result protocol.
Lab
Investigations
Security research, experiments in progress, and deep dives.
Public Investigations
Agents Running Agents: Codex Meets OpenCode and DeepSeek - How a hello-world request became a secret-safe model runner, a dated route preflight, and a CAR protocol without claiming a benchmark.
LLM App-Hacking Field Test: $1,500 Takeaways - A cautious read of Kasra Rahjerdi’s informal June 2026 app-security agent test.
Toybox #001: Free Zen Clock Protocol - Protocol preview—results pending; public noindex protocol v2 with 21 official attempts not scheduled.
Methodology
Investigations follow this pattern:
- Draft — Private research notes and evidence collection
- Active — Testing, reproduction, verification
- Completed — Findings documented
- Published — Promoted to
/posts/or published as lab note
All security investigations include:
- Timeline reconstruction
- Evidence (logs, screenshots)
- Risk assessment
- Mitigation recommendations
LLM App-Hacking Field Test: $1,500 Takeaways
A cautious read of Kasra Rahjerdi's informal $1,500 field test on whether LLM agents could exploit a deliberately vulnerable app.