Gemini Managed Agents: 3 of 5 Sandbox Settings Send Your GitHub Token on Every Call
Last Updated on October 29, 2026 by Editorial Team
Author(s): Chew Loong Nian – AI ENGINEER
Originally published on Towards AI.
Gemini Managed Agents: 3 of 5 Sandbox Settings Send Your GitHub Token on Every Call
Google’s talk says secrets never enter the sandbox, using the credential route on its slide. Three of the other four routes in the SDK’s environment config put the token in every request body, and in the debug log when it is on. A 73-line script shows the difference without an API key.
After reviewing the talk’s claim about storing tokens with client.credentials.create() and having a proxy inject them on the wire, the article tests how google-genai 2.33.0 actually handles tokens across five “environment” configuration routes. It describes five input fields (env, env with value, header transform, credential, and env with credential) and shows that three “plain” routes embed the real GitHub token in every request and are repeatedly exposed in the SDK debug logs when enabled, while the credential-based routes keep the real token out of the sandbox and send it only once during credentials setup. The author then provides a self-contained local script that uses a fake API server plus GOOGLE_GENAI_DEBUG=1 to count token occurrences in raw request bytes and logs across repeated interactions, reports the resulting counts, explains the debug-log risks (including how it can leak both the API key and token if captured), and concludes with the practical fix: create the credential once outside the agent-running code and reference its credential ID via the allowlist, using trusted_domains appropriately.
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ AI Engineering For Production — our new book, coming before the end of 2026. Join the waitlist.
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
![[Day 37/100] Build Your Own Agent Memory Layer with Postgres and pgvector](https://miro.medium.com/v2/resize:fit:700/1*-Doos2nmUT3FPZ-Qxbk3Hw.png "[Day 37/100] Build Your Own Agent Memory Layer with Postgres and pgvector")
