Author: Alan Blount (@zeroasterisk) Date: June 29, 2026 Context: Evaluating sandbox providers for the Flue agent framework with focus on GEAP (Gemini Enterprise Agent Platform) integration
Flue supports 10+ sandbox providers via its SandboxApi interface. This report compares the five most significant providers across capabilities that matter for AI agent workloads. GEAP is the only provider with enterprise governance built in, but trades developer experience for security posture.
Scion Hub Friction Log & Architectural Review
Scion Hub Friction Log & Architectural Review
Author: Alan Blount (@zeroasterisk) Date: June 2026 Context: Setting up Scion Hub on a personal NAS, with assistance from Antigravity (Gemini agent)
Google's Agent Development Kit (ADK, google/adk-python) currently pins a2a-sdk>=0.3.4,<0.4.0 and uses Pydantic-based types for all A2A protocol objects. The a2a-sdk v1.0 release migrates every protocol type from Pydantic models to Protobuf-generated classes, which breaks construction patterns (Part(root=TextPart(...)) becomes Part(text=...)), enum naming (TaskState.working becomes TASK_STATE_WORKING), field access (.root.text becomes .text), and serialization calls (.model_dump() is gone on protobuf objects). The AgentCard.url field is removed in favor of supported_interfaces.
ADK's A2A integration is marked experimental and spans approximately 20 source files under src/google/adk/a2a/ plus src/google/adk/agents/remote_a2a_agent.py, with roughly 12,000 lines of test coverage across 15+ test files. The migration affects every file in this surface area. The v1.0 SDK ships a a2a.compat.v0_3 compatibility layer, but it does not cove
An argument stitched from Solve Everything: Achieving Abundance by 2035 (Alexander D. Wissner-Gross & Peter H. Diamandis).
Three moves: bet on the open rail; standardize the target; run the target over the rail.
Raw PyPI download counts conflate humans with mirrors, CI/CD, and transitive
dependencies. When a package "takes off" or two packages look like a migration,
the naive read is usually wrong. This is a method (+ copy-paste SQL) to tell a
human adoption wave from a single-image automation fleet, using the public
BigQuery table bigquery-public-data.pypi.file_downloads.
How Agent Finder / AI Catalog (discovery + semantic search) and the Agent Name Service (ANS) (domain-anchored identity + verification) fit together — written for the people who maintain each.
One-line thesis: they are different layers and they stack. Agent Finder answers "what agents exist and which one fits my need?" ANS answers "who is this agent, and is it provably the real one?" The highest-value work is wiring them at one seam — search semantically, verify cryptographically.
ANS already designs for this: its architecture **decouples identity (the Registration Authority) from
A review of the openclaw/openclaw repository reveals numerous open issues related to the existing google-vertex provider (e.g., #48033, #49039, #49191, #56253).
The Core Bug: The new @google/genai SDK is used downstream (via @mariozechner/pi-ai). Currently, when Application Default Credentials (ADC) are detected, a sentinel value ("<authenticated>") is returned. OpenClaw blindly treats this string as a literal API key and passes it down. The @google/genai SDK constructor sees a truthy apiKey, assumes it should use the Gemini Developer API, and sends x-goog-api-key: <authenticated>, which immediately results in a 401 UNAUTHENTICATED from Vertex AI.
There is a clear need for a unified google-genai provider that gracefully handles both standard API keys and IAM/Vertex ADC workflows, mapping cleanly to the @google/genai SDK's design.