AI Dev Gems: Week of Oct 17–23, 2026
OpenAI's open-source application security agent, shipped as a CLI and TypeScript SDK, that finds, validates and proposes fixes for vulnerabilities in your codebase.
Why: Official OpenAI repo (openai/codex-security) with npm releases; npm-v0.8.2 was published Oct 29, 2026. A first-party security agent you can script into CI.
Zero-dependency C++23 CLI and MCP server that hands coding agents a ranked call-graph map of a repo, then checks their edits for blast radius, tests to run and quality deltas.
Why: Red Hat Emerging Technologies project (redhat-et/ripwire) with frequent tagged releases; v0.6.9 shipped Oct 29, 2026. Gives agents context and verification at once.
Google's official set of modular SKILL.md files that teach coding agents Android development best practices, from Compose UI to build setup.
Why: Maintained under the official android GitHub org (android/skills); release v1.0.19 landed Oct 29, 2026, showing steady versioned updates.
NVIDIA-verified, signed Agent Skills that install into Claude Code, Codex and other coding agents to drive CUDA, RAG, simulation and robotics workflows.
Why: Official NVIDIA/skills repo, created Feb 2026 and still under active commits as of Oct 30, 2026; skills are signed and verified by NVIDIA.
CPU-only code search library and MCP server that returns exact code snippets to coding agents, with no API keys or GPU required.
Why: From MinishLab (makers of Model2Vec); the repo says it uses 99% fewer tokens than grep plus read, and v0.7.1 was released Oct 29, 2026.
Stateless command-line interface from Google that lets coding agents and shell scripts call Gemini models, media generation and managed agents without writing SDK code.
Why: Official google-gemini/gemini-api-cli repo, marked experimental; v0.4.5 was published Oct 30, 2026.
AWS Labs' harness-neutral, gated workflow engine that turns AI coding assistants into structured, auditable software-delivery pipelines.
Why: Open source at awslabs/aidlc-workflows with near-daily preview releases; v2.11.3-preview.20261030.1 shipped Oct 30, 2026.
Microsoft research-preview platform for measuring how well AI coding agents handle your product, using reusable criteria and repeatable experiments.
Why: Lives at microsoft/scope (created July 2026) and is labeled an agentic experience evaluation platform in research preview; commits continued through Oct 30, 2026.
Pytest-native framework from Microsoft for writing safety and security tests against agentic AI applications alongside your normal test suite.
Why: Open source at microsoft/RAMPART; first tagged release v0.1.0 in May 2026, with commits through Oct 29, 2026.
GitHub Next's real-time multiplayer planning environment where engineers and AI agents co-write implementation plans before code is written.
Why: GitHub Next project (githubnext/chopin) shipping a CLI; cli-v0.0.21 was released Oct 22, 2026.
Open-source, cross-harness memory layer that records coding-agent sessions and turns what worked into reusable skills for future runs.
Why: Asymptote-Labs/agent-beacon ships rapid releases; v1.3.51 was published Oct 30, 2026.
Git-native work board and CLI that coordinates coding agents from approved specs, where a second agent must verify each task before it can merge.
Why: pullboard-dev/pullboard repo created Oct 2026; v0.9.3 was released Oct 27, 2026.
Hugging Face's private cloud workspace for running and managing Claude Code, Codex, Gemini CLI and other coding agents from a browser, deployable as a Space.
Why: Official huggingface/agent-manager repo with a duplicable Space template; active commits through Oct 30, 2026.
Open-source CLI that lets Claude Code or Codex work next to real secrets while the model only ever sees masked placeholders.
Why: ManduSec/mandu was created Oct 19, 2026 and is already at v0.2.0-beta.8 (released Oct 22, 2026); an early but focused fix for secret leakage to agents.