Observed Signal · Jul 22, 2026 · Technical Release · Source: DEV Community · Impact: 3/5 · Sentiment: Positive
Stateful Image-Editing Claude Code Skill Using Gemini
A third-party open-source project, nb2lite-skill-claude, packages Google’s gemini-3.1-flash-lite-image model behind a small FastMCP server and a Claude Code skill to enable stateful image generation and iterative editing. The setup uses Google’s Interactions API so each generation returns an interaction_id that preserves visual context on the server, allowing incremental edits (e.g., “add a neon sign”) without re-describing the entire scene. The repo provides an MCP server exposing four tools (generate_image, edit_image, edit_local_image, get_help), installation options (Claude plugin marketplace, repo clone, project install, Docker), and ships under an Apache-2.0 license. The project dogfoods itself: the article’s cover was generated by the skill. Publication date: 2026-07-22.
Open-source integration demonstrating stateful image editing with Google’s Interactions API and an MCP server; useful for creative workflows and assistant-tool integrations but not a platform-level policy or major vendor product announcement.
Track Google Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- The GitHub repo nb2lite-skill-claude wraps gemini-3.1-flash-lite-image in a FastMCP server and a Claude Code skill (Apache-2.0).
- The project uses Google’s Interactions API to provide stateful image editing via persistent interaction_id handles.
- The nb2lite-agent MCP server exposes four tools: generate_image, edit_image, edit_local_image, and get_help.
- A Docker image xbill9/nb2lite-agent is published for running the MCP server in a container.
- The article and repo were published on 2026-07-22.
Connected Companies & Entities
3 Entities mapped“Google's Nano Banana 2 Lite — the friendly nickname for `gemini-3.1-flash-lite-image` — takes a different approach....”
“This is a third-party community project, not affiliated with or endorsed by Anthropic or Google....”
“Repo: github.com/xbill9/nb2lite-skill-claude (Apache-2.0)...”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Stateful Video Editing Skill Using Gemini Interactions
A third-party open-source project (omni-skill-claude) packages Google’s gemini-omni-flash-preview model (Omni Flash) behind a small FastMCP server and a Claude Code skill to enable stateful, iterative video generation and editing. The setup exposes eight MCP tools (generate, edit, animate, interpolate, subject-driven generation, restyle user videos, upload to YouTube, and help) and relies on Gemini’s Interactions API to persist visual context with interaction IDs so subsequent edits preserve continuity. The repo is on GitHub (Apache-2.0) and requires Python 3.10+, a Gemini API key, and Claude Code; video generation is synchronous, billable, and supports inline or File-API delivery modes depending on size.
Guide: How to Build and Optimize Claude Skills
This guide explains how to build, test and optimize Claude Skills — permanent, reusable instruction files that automate tasks for Anthropic's Claude models. A Skill is a local folder containing a case-sensitive SKILL.md (with YAML frontmatter) and optional references/scripts; folders use kebab-case and are placed in ~/.claude/skills/ so Claude can auto-detect them. The guide covers writing aggressive YAML trigger descriptions, defining precise triggers and quality standards, workflow structure, edge-case handling, using scripts for precise computation, and handover patterns for session continuity. It also describes Skills 2.0 capabilities — evaluation frameworks, A/B testing, and automated description optimization — plus a meta-skill called skill-creator that can generate, evaluate and benchmark Skills (including tests that compare a Skill against raw Claude). The piece emphasizes iterative testing and clear non-overlapping Skill territories.
Agent-built generative video pipeline using Claude Code
A developer describes building a two-minute video entirely via an agentic Claude Code session (named “Simona”) that created and composed image generation, text-to-speech, AI-video, and ffmpeg editing skills. The post is a technical walkthrough showing how the agent iteratively built reusable "skills" (with SKILL.md docs and CLI wrappers), tracked costs in a WORKLOG.md ledger, and recovered after a git mishap that deleted assets. The author lists the models and services used (OpenAI gpt-image-2, Google Gemini/Nano Banana, Seedance 2.0, Kling, LTX, ElevenLabs, Google TTS, local Kokoro), provides a cost breakdown ($27.76 for the final locked cut; $45.26 total project spend), and documents engineering patterns and guardrails for safe agent-driven media production.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.