Gists

Have you ever had a brilliant solution get completely crushed by a single comment on a technical blog post?

Just last week, on July 8, 2026, I published a post introducing the Pyramid Aggregator. It was a beautiful story of technical lineage: we took a string summation algorithm originally designed on October 13, 2016 (to optimize Google Apps Script) and applied it to modern Large Language Models (LLMs). By organizing document merging into a balanced, parallel tree topology, we successfully bypassed the AI’s “Lost in the Middle” bias and achieved a 95% execution speedup using the Antigravity Python SDK.

Gists

Have you ever tried to read a massive pile of reports and summarize them in under 50 words? It’s hard. Now, imagine asking a cutting-edge Large Language Model (LLM)—like Gemini—to do it.

You might think AIs have perfect memories, but they don’t. When forced to aggregate information from dozens of documents under strict length constraints, AIs suffer from severe “memory loss” biases. They either ignore the middle of your documents or completely forget the older information they read first.

Gists

Abstract

Executing autonomous AI agent payloads in Google Workspace via the Apps Script API’s scripts.run method introduces severe security risks. This article presents a novel sandboxing proposal designed specifically for the scripts.run method, using ggsrun as the orchestrator to execute code safely and efficiently. By performing in-memory token replacement and uploading a separate, alphabetically-prioritized guard file, this approach achieves robust API-level containment. Guided by ggsrun’s automated backup and default rollback lifecycle (exe1), the remote environment is immediately restored, providing a clean, dependency-free security model for AI-driven Workspace automation.

Gists

Motivation

To be quite honest, “Hooks”—the shell commands we trigger at specific points when generative AI agents process tasks—were something I used blindly for a long time. Whenever colleagues asked me about them, I realized I lacked any real confidence in explaining how they actually work. However, when I migrated from Gemini CLI to the new Antigravity CLI, I noticed that the hooks system carried over. This felt like the right moment to stop guessing and finally develop a precise, deep understanding of the mechanism. I went back to the basics to analyze exactly how hooks operate under the hood and how we can use them effectively in the Antigravity environment. My goal is to demystify hooks so we can write them with confidence, and if this guide proves useful to your own workflows as well, I would be very glad.

ggsrun was updated to v5.3.7.

• v5.3.7 (June 2026) - Simplified Quick Onboarding, On-demand Setup Prompting, Optional Credentials Path, and Seamless Configuration Initializer Introduced the groundbreaking, extremely easy-to-use $ ggsrun setup onboarding command to dramatically simplify Google Cloud API and OAuth2 credentials setup, while keeping traditional $ ggsrun auth fully intact for backward compatibility. This command automates Google Cloud Workspace API enablement (Drive, Sheets, Slides, Docs, Google Apps Script, Gmail) using GCP Quick Flows, immediately redirecting users straight to the Credentials creation page. It removes any credential filename renaming constraints (you can load credentials from any custom file path like {your path}/{credential name}.json rather than renaming to exactly client_secret.json) or supports manual credential copy-pasting. Added a dynamic config initializer (ggsrunIniForSetup) to gracefully bypass “missing client_secret.json” startup errors on first-time runs, and added interactive prompts to configure Script IDs and Web Apps URLs with automated pre-filled defaults from existing configurations.

You can check ggsrun at https://github.com/tanaikech/ggsrun.

GASADK was updated to v2.0.0

• v2.0.0 (June 24, 2026)
  • Exposed Google Workspace APIs as a Dynamic MCP/A2A Server.
  • Added a standalone deployment script (DeployMcpServer.js) under samples/googleapiapp-mcp-server to expose the GoogleApiApp library.
  • Created a detailed beginner-friendly guide for configuring and deploying Google Workspace services (Sheets, Drive, Docs, Calendar, Gmail, etc.) as dynamic MCP tools.
  • Added API Quota Safeguard: Track model token consumption per session and enforce session-level token usage limit check via maxTokensPerSession configuration.
  • Added Hook & Session Propagation: Automatically propagate parent hooks security context (hookManager and sessionId) and token constraints down to child sub-agents during hierarchical delegation.
  • Added Human-in-the-Loop (HITL) Integration: Supported decision: "suspend" decisions in BeforeTool hooks. Seamlessly serializes and stores task state to PropertiesService (under HITL_STATE_[sessionId]) and throws a "SUSPENDED" exception. Added saveState(), loadState(sessionId), and resume(sessionId, decision, approvedArgs) methods to recover and run suspended agent execution loops.
  • Synchronized src/lib/ Core Libraries: Upgraded bundled core components (GeminiWithFiles.js, A2AApp.js, MCPApp.js, MCPA2Aserver.js, and FileSearchApp.js) to support the GasHookManager hooks system natively (allowing interceptors like BeforeModel, AfterModel, BeforeTool, and AfterTool).
  • Ecosystem Modernization: Upgraded default model configurations to models/gemini-3.1-flash-lite across all workspace libraries, tests, and examples to provide modern, cost-effective defaults.
  • Modular Standalone Twin Synchronization: Propagated and fully published these updates to all five standalone local repositories (GeminiWithFiles_GAS_lib, A2AApp_GAS_lib, MCPApp_GAS_lib, MCPA2Aserver_GAS_lib, FileSearchApp_GAS_lib) with next-version tags.

You can see the detail information here https://github.com/tanaikech/adk-gas

ggsrun was updated to v5.3.3.

• v5.3.3 (June 2026) - Recursive Directory Walk, Safe Interactivity & GAS Zip Download Enhanced updateproject (alias ud) command to recursively traverse folders specified via -f / --filename to batch overwrite remote GAS projects. Prints targeted local files in a beautiful bullet list using pterm.BulletListPrinter and requires explicit interactive confirmation (Y/N) in CLI/TUI modes before mutating Google Drive files. Supported downloading whole Apps Script projects directly as local packaged .zip archives via ggsrun download -i <fileId> -z. Added robust security warnings to the updateproject MCP tool description directing AI agents to obtain user approval before calling the tool. Introduced a complete automated integration testing suite (cli_test.go) validating download structures, document conversions, standalone uploads, and binary fallbacks.
• v5.3.2 (June 2026) - Script Upload Flag Registration and TUI Focus Fallbacks Fixed a TUI upload crash where converting and uploading .js/.gs files to standalone Apps Script projects threw a panic: internal process exited with code 1 due to unregistered projectname and googledocname flags in createOpContext. Integrated full text file previews on Enter for remote files, and implemented dynamic importFormats MIME type lookup via utl.GetImportTargets to automatically bypass conversion prompts for unconvertible file types, as well as robust focus restoration.
• v5.3.1 (June 2026) - Script Upload Routing Fixes, Non-Convertible Upload Fallbacks, and TUI Error Propagation Fixed a bug in concurrentUpload that caused .js/.gs/.gas uploads to throw HTTP 400 Bad Request by redirecting script project uploads to the Apps Script project builder and overriding raw script uploads to text/plain. Allowed non-Workspace files (such as .zip or .mp3) to bypass conversion checks and upload as-is when no conversion format is requested. Integrated TransferResult and FileInf error inspections in the TUI filer (ggsrun fd) to propagate upload/download failures to the user instead of failing silently.

You can check ggsrun at https://github.com/tanaikech/ggsrun.

ggsrun was updated to v5.3.0.

• v5.3.0 (June 2026) - Responsive TUI Filer (FD Mode) Enhancements, Focus Persistence, and Platform Compatibility Fixes Refactored TUI Filer (FD Mode) popup layouts using tview.Flex to center dialogs (errors, sorting, details, help) and prevent text clipping. Implemented focus locking to preserve active panel focus post-action. Added wrap-around to lists and mapped the y key to yank (copy) selected file paths or File IDs to the clipboard. Resolved compilation errors on 32-bit Linux platforms (e.g. linux/arm) by explicitly casting syscall.Stat_t Ctim fields to int64 inside build-tagged files, and adapted the mock simulation test suite (fd_test.go) to match the new responsive structures.
• v5.2.4 (June 2026) - Latest MIME Type Formats, CLI Option Help Details, Concurrent Conversion Overhaul, and Destination Directory Support Updated internal MIME type mapping definitions (googlemimetypes.go) to synchronize with the latest Google Drive API importFormats/exportFormats. Revamped the CLI options help display for --extension (download/revision) and --convertto (upload) to explicitly list all supported formats. Overhauled the concurrent upload/download engines to handle --convertto / --noconvert directly in parallel streams without falling back to the legacy single-threaded uploader, adding validation capability checks and graceful error warning feedback. Added --destination (-d) option to the download and revision commands to allow specifying the target local directory for saving downloaded files.
• v5.2.3 (June 2026) - Directory Reuse Conflict Resolution, Output Control, and CLI/MCP Alignment Upgraded the directory upload conflict resolution mechanism to silently and recursively reuse existing remote folders without prompting. Aligned the --conflict-mode behavior for -j / --jsonparser CLI runs to match the automated MCP mode (defaulting to OverwriteIfNewer, overridable via --cm). Muted TUI output and progress bars (mpb) when running with the -j option to return clean JSON. Supported --cm as a shorthand alias for --conflict-mode in file transfers.
• v5.2.2 (June 2026) - MCP Help Display Expansion, Safety Review Prompt, Dual-Mode Conflict Engine, and File-Level Error Feedback Expanded ggsrun mcp -h (and --help) to display all exposed MCP tool names and their detailed description outputs directly. Implemented strict programmatic safety review prompts inside the exe1 MCP tool description, instructing LLMs to statically analyze Apps Script payloads for API mutations (write/update/delete) and obtain user Y/N confirmations before running, while allowing read-only scripts to run automatically. Re-designed the conflict resolution engine into a dual-mode system: automated and non-interactive for MCP server sessions (defaulting to OverwriteIfNewer, with options for Ignore and Rename), and preserving legacy interactive CLI prompts for raw executions. Refactored parallel transfer loops to capture and return detailed file-level error feedback instead of crashing.
• v5.2.1 (June 2026) - Dynamic CLI Help Customization, Beacon Script Integration, and Namespace Binding Updated the CLI help systems for e1, e2, and w to integrate comprehensive execution command examples (including stateless beacon checks) dynamically within both the --help flag screens and optionless execution error overlays. Fixed a namespace bug where evaluated scripts executing ggsrunif.Beacon() inside the library threw a ggsrunif is not defined ReferenceError, by binding ggsrunif to the library’s global execution context.
• v5.2.0 (June 2026) - Go standard layout, WSL2 browser integration, Web Apps URL registration, CLI UX hardening, and MCP Server Schema Improvements Reorganized the codebase to follow the standard Go project structure (main.go, /internal/app/, /internal/utl/). Expanded ggsrun auth to request Web Apps URL registration and dynamically persist it in ggsrun.cfg, allowing ggsrun w to run without the -u option. Integrated WSL 2 environment detection to prompt the user to choose between the Windows host browser, WSL/Ubuntu native browser, or manual URL copy-pasting. Upgraded ggsrun e1, ggsrun e2, and ggsrun w commands to dynamically print full CLI flag helps alongside custom usage examples. Improved the MCP server (ggsrun mcp) tools schema, adding rich parameter descriptions, Drive API query examples, new scriptfile/stringscript parameters to the exe1 schema, searchbyid parameter to the filelist schema, and making scriptid optional by resolving automatically from ggsrun.cfg (via GGSRUN_CFG_PATH or the local directory). Refined tools/call backend handling to safely strip null/empty values.
• v5.1.1 (May 2026) - Modular Handlers & Enhanced MCP Server Core Refactored the codebase to modularize legacy single-file command handlers into dedicated, organized handler files (handler_download.go, handler_upload.go, handler_transfer.go, handler_mcp.go, handler_execute.go). Strengthened the MCP server core (ggsrun mcp) by capturing stdout and stderr execution logs for comprehensive error recovery. Embedded full support for --conflict-mode inside the MCP JSON-RPC schemas and standardized file transfer outputs into TransferResult to support interactive multi-turn collision resolution in LLM conversations. Fully updated pre-built binaries for all major architectures.
• v5.1.0 (May 2026) - Advanced Conflict Resolution Engine Introduced a robust pre-computation conflict resolution matrix for both download and upload commands via the new --conflict-mode (-cm) flag. Users can now choose from skip, overwrite, rename (appends timestamp _YYYYMMDD_HHMMSS to avoid collisions), or update (syncs only if the source file is newer than the target). Includes interactive fallback CLI prompts if no mode is specified. Deprecated the legacy --overwrite (-o) and --skip (-s) options in favor of --conflict-mode. To avoid Drive API rate limits during massive concurrent uploads, metadata query is pre-fetched in bulk.
• v5.0.3 (May 2026) - CLI UX Overhaul & Dynamic TUI Integration Introduced a highly visual, modern Terminal UI (TUI) powered by pterm for exe1, exe2, and webapps commands. Added interactive loading spinners with anti-ghosting fixed-width padding (%-70s) and beautifully structured execution reports. Maintained strict backward compatibility by preserving pure JSON output streams via the -j flag for CI/CD pipeline automation.
• v5.0.2 (May 2026) - Secure Web Apps Protocol Upgrade Upgraded the webapps command to natively support “Only myself” execution deployments by bridging OAuth tokens (drive scope) across Google’s HTTP 302 Auth Redirects. Ported the IIFE/JSON-literal double-eval protections from exe2 to webapps.
• v5.0.1 (May 2026) - Execution Engine Hardening & Double-Eval Eradication Eliminated the V8 engine double-eval 500 server crash during dynamic script execution by enforcing IIFE and JSON-literal payload encoding. Redefined -f flag mapping for proper API gateway resolution in exe2. Added precision deployment documentation for stateful and stateless execution modes.

You can check ggsrun at https://github.com/tanaikech/ggsrun.

Gists

Abstract

This article explores the integration of Google Workspace with the Antigravity CLI, the high-performance successor to the legacy Gemini CLI. This integration is critical because it bridges the gap between low-latency, local agent execution and cloud-native enterprise productivity platforms. We demonstrate this framework by evaluating five core developer tools—the Google Workspace CLI, gas-fakes, ggsrun, GASADK/GoogleApiApp, and goodls—and mapping their capabilities into distinct local, hybrid, and cloud execution layers. Our analysis reveals how this unified architecture streamlines complex, multi-step agentic workflows while optimizing resource consumption, establishing a blueprint for next-generation workspace automation.

Gists

Abstract

This paper presents a serverless architecture that overcomes the stateless nature and 6-minute execution limit of Google Apps Script (GAS). By configuring a 1-second immediate timeout in UrlFetchApp loopback calls, an orchestrator dispatches background tasks and terminates immediately. This design frees up the caller’s execution quota while the target Web App runs to completion in an isolated container. Combined with a transactional Google Sheets state machine, this design supports self-perpetuating parallel MapReduce runs and multi-turn, state-hydrated generative AI agent networks without external compute infrastructure.