Initialize an Agent Harness with Forge CLI
Spec document ingestion and merge-safe CLAUDE.md handling show thoughtful DX, but agent scaffolding is crowded.
A terminal bridge to Anvil apps via the Server Uplink. Run diagnostics, query Data Tables, invoke server functions, and drop into an interactive REPL — all from your shell or an AI coding assistant.
Agent-safe key handling prevents credential leaks in AI coding agent transcripts.
Anvil.works developers, teams using AI coding agents
Official Anvil CLI · Database CLI tools
anvil-uplink-cli is a thin CLI over that library. The bits worth looking at:
- query, call, row, run, repl — the things you'd otherwise do by pasting into the web Server Console - multi-app profiles, pretty/JSON output, works the same on Windows / macOS / Linux / WSL - a dotenv: key_ref scheme designed for AI coding agents: the Uplink key stays in a gitignored .env, never in a tool-call argument, never in the agent's context
That last bit is the part I most want feedback on. The standard "just use an env var" advice works fine for humans, but when a coding agent is driving your CLI, any inline KEY=... cmd lands in the transcript — and a malicious row from the database could then instruct the agent to exfiltrate it. dotenv: side-steps this by making the CLI read the key itself and never exposing it to the caller.
Built this scratching my own itch while using Claude Code on my Anvil apps. Early alpha, MIT.
Spec document ingestion and merge-safe CLAUDE.md handling show thoughtful DX, but agent scaffolding is crowded.
Unified CLI for coding agents, but shell aliases or a simple script could do the same.
Actual Kubernetes operator for agent lifecycle, but orchestrating agents is still a niche use case.
Plumbs Cursor/Claude Code into Slack without rebuilding the agent logic.
Credential proxying keeps keys out of sandboxes, unlike Ramp Inspect.
Yet another Docker wrapper for AI agents, but drops all Linux capabilities by default.