AI Dev Gems: Week of Nov 7–13, 2026
Rank 1Poki
Let AI agents operate a real iPhone from a Mac, with no jailbreak or extra hardware, through an MCP server.
What makes it a gem: Added Nov 7. The GitHub repo was created Nov 5 and already has 1,087 stars. Agents read the iPhone's accessibility tree and send native touch events. A 0.4.0 beta landed Nov 13. Tags: #ios #mcp
Rank 2Elephant-Goldfish Model
Draft software specs with a context-rich agent, then review them with zero-context subagents that don't share the earlier conversation.
What makes it a gem: Added Nov 7. Google open-sourced the plugin in October; Google Cloud Tech posted about it this week, and the repo has 18,622 stars. A context-rich Elephant session drafts the spec; Goldfish subagents review it without the earlier conversation. Tags: #agent-skills #specifications
Rank 3ttsc
Build, type-check, lint, and run TypeScript using a shared TypeScript-Go compiler.
What makes it a gem: New to the catalog Nov 13. Its compiler graph gives coding agents access to resolved symbols over MCP. The same compiler reports type errors and lint findings. Tags: #typescript #mcp
Rank 4sandbox-mcp
Give coding agents a fresh, disposable Windows 11 PC through Windows Sandbox, as an MCP server and a CLI.
What makes it a gem: Added Nov 8. Scott Hanselman's tool boots a throwaway Windows 11 machine so an agent can install an app, drive the UI, and keep the screenshots after the guest is deleted. The repo has 426 stars and was created Nov 6. Tags: #sandbox #mcp #windows
Rank 5MergeAtlas
Turn a pull request into an interactive local map of why files changed and what they affect.
What makes it a gem: Added Nov 12. The map gives reviewers a way to follow changes across files before digging into individual lines. Tags: #code-review #agent-skills
Rank 6ThreadPort
Package unfinished coding agent work with the evidence needed to pick it up in another session.
What makes it a gem: Added Nov 12. It gives the next agent a record of the work in progress alongside the handoff. Tags: #coding-agents #handoffs
Rank 7forked
Branch and rewind an AI agent's saved state with an action log that can replay recorded results.
What makes it a gem: Added Nov 7. You can retry from a checkpoint and use the recorded result for an action that already ran. Tags: #agent-infrastructure #debugging
Rank 8token-max
Replay Claude Code and Codex transcripts against API prices to compare what different settings would cost.
What makes it a gem: Added Nov 12. It starts from your own agent sessions, so the estimate reflects how you work. Tags: #coding-agents #costs
Rank 9Bend
A Python-like language that blocks incorrect AI-written code with compiler-checked proofs and automatic CPU and GPU parallelism.
What makes it a gem: A carry-over: added to the catalog Oct 1. The official repo is github.com/bendlang/bend (github.com/HigherOrderCO/Bend redirects there) and has 25,031 stars. Commits continued through Nov 13, and the latest release is Bend 2.0.40, published Nov 5. Tags: #languages #verification
Rank 10airgap-check
Scan an AI codebase for network calls, telemetry, and downloads that can break offline deployments.
What makes it a gem: Added Nov 12. It looks for first-run downloads that can slip through when you test with internet access. Tags: #offline-ai #static-analysis
Rank 11DiffNotes
Add notes, screenshots, and live preview links to specific changes in a pull request.
What makes it a gem: Added Nov 7. Reviewers can open a preview at the part of the app a note describes. Tags: #code-review #pull-requests
Rank 12Consumer Agent Testing Tool (CATT)
Test how AI assistants use your MCP server, app, website, or API.
What makes it a gem: Added Nov 7. Tests can check what actually changed on the service side, even when the assistant says it succeeded. Tags: #testing #mcp
Rank 13Bruin MCP
Give coding agents access to Bruin's data pipeline tools through MCP.
What makes it a gem: Added Nov 7. The server runs from the Bruin CLI, so an agent can work with the sources and pipelines already configured in the project. Tags: #data-engineering #mcp
Rank 14Knowledge Work SDK
Build Python agents that check research and analysis against a rubric and revise when they fall short.
What makes it a gem: Added Nov 12. The SDK exposes the rubric and lets developers supply their own criteria for a task. Tags: #agent-frameworks #evaluation