what is resale-agent-skill-hub fr?

SkillHub is a Python toolkit that turns Claude into an assistant for second-hand resale across consumer-to-consumer marketplaces like eBay and Depop. It ships as eight Claude Code skills plus an MCP server. Once the server is wired up to an MCP-compatible client such as Claude Desktop, Cursor, or VS Code, the assistant can call seven tools that together cover photo recognition, pricing, account login, listing creation, repricing schedules, and automatic delisting once an item sells. The core flow is a seven-step pipeline. broker-recognize takes a product photo and uses Claude Vision via the Anthropic API to produce structured product info. broker-auth opens a real browser through Playwright so the user logs in to each platform once, and the session cookies are persisted for about 30 days. broker-price searches the platforms while logged in and produces a transparency report. broker-fuse sets a floor price that every later price-mutating action must respect. broker-card generates the listing card, validates the floor, asks the user to confirm, then auto-publishes. broker-schedule registers a launchd or cron job for daily repricing checks. broker-delist auto-detects a sale on one platform and removes the listing from all others without further confirmation. A shared engine in the broker_core package handles the session manager, the platform clients, the listing state file at ~/.broker/listings.json, the launchd or cron registration, and an append-only audit log at ~/.broker/audit.jsonl. Every pricing decision, floor check, publish, reprice, and delist event is written to that log so the user can audit what the agent did and why. By default the project runs in mock mode, with BROKER_MOCK_MODE set to true, which returns simulated data for searches, publishes, and delists. That lets a new user try the full pipeline end-to-end without any platform accounts or real API calls. Switching BROKER_MOCK_MODE to false and running the broker-auth script opens the browser flow that binds real accounts. Installation is a git clone, pip install -e ., and pip install playwright followed by playwright install chromium. The recommended setup path is to point Claude Desktop at broker_core/mcp_server.py via the MCP servers config file. Requires Python 3.9 or newer and an ANTHROPIC_API_KEY for photo recognition. The license is MIT.