agy-delegate

Antigravity Delegate

You are the orchestrator. This skill lets you hand a bounded coding task to a separate implementer - the Google Antigravity CLI (agy) - then review what it produced and land it yourself. You write the brief and own the judgment; Antigravity does the typing in its own conversation; you verify and commit.

There are two dispatch paths and the loop around them is identical:

• MCP (preferred) - the claude-agy-mcp bridge exposes delegate, follow_up, adversarial_review, delegate_many and friends as tools. Use this whenever the tools are available.
• CLI relay (fallback) - scripts/relay.mjs wraps agy --print and writes a result.json. Use it when the MCP server is not registered, so an orchestrator that can only run a shell command and read a file can still drive the loop.

When NOT to use this

• The task is small enough to just do inline - delegation overhead is not worth it.
• Neither the MCP server nor the agy CLI is available and authenticated.
• You want to write the code yourself, or you only need Antigravity's opinion on code you wrote - a read-only delegate or adversarial_review covers that, but a plain review may not need delegation at all.

Prerequisites (check once)

1. Load the tools. In an orchestrator where the bridge's tools are deferred, load them in a single call, not one per tool:

   ToolSearch("select:mcp__claude-agy-mcp__delegate,mcp__claude-agy-mcp__follow_up,mcp__claude-agy-mcp__adversarial_review,mcp__claude-agy-mcp__delegate_many,mcp__claude-agy-mcp__agy_status,mcp__claude-agy-mcp__set_model")
   

2. Call agy_status. It never reaches agy, so it is a free liveness check. It reports the agy version, the models on offer, the resolved model chain per tool, quota cooldowns, in-flight runs, and whether a set_model choice has already been recorded.

3. Settle the model once, if agy_status shows no choice recorded. The bridge defaults to AGY_ASK_MODEL=true and refuses to delegate until set_model has been called - the first delegation returns an error, not a result. Ask the user "proceed with the default, or change model or effort?", then call set_model once: no arguments accepts the default (Gemini Flash High), or pass what they chose. The choice is saved per machine and no tool asks again. An explicit model on a single call still wins for that call.