Summary

• Fixes API validation failures when switching AI providers mid-session
• Normalizes tool call block formats (toolUse, toolCall, functionCall) to canonical shape before sending to any provider
• Prevents "input: Field required" errors when provider serializers encounter mismatched formats

Problem

When users switch providers mid-session (e.g., from Google Antigravity to Anthropic), the session history may contain tool call blocks in different formats:

• Anthropic / pi-ai: { type: "toolCall", arguments: {} }
• Google Antigravity: { type: "toolUse", input: {} }

This mismatch causes API validation failures.

Solution

Adds normalizeToolCallBlocks() to the session sanitization pipeline, converting all tool call variants to the canonical { type: "toolCall", arguments: {} } shape.

Test plan

• Switch from Google Antigravity to Anthropic mid-session with tool calls in history
• Verify no validation errors occur
• Confirm tool call history is preserved and functional

🤖 Generated with Claude Code

Greptile Overview

Greptile Summary

This PR appears to normalize tool-call blocks in the Pi embedded runner's Google provider integration so that session history remains compatible when switching between AI providers mid-session. The change introduces (or wires in) a normalization step that converts provider-specific tool call variants (e.g., toolUse with input) into a canonical shape (toolCall with arguments) before sending history to downstream providers, preventing schema validation errors like input: Field required.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk.
• The change is scoped to session history normalization for tool call blocks and is intended to improve cross-provider compatibility without altering core control flow. No obvious security- or correctness-critical issues were identified from the provided context.
• src/agents/pi-embedded-runner/google.ts

_{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!}

Context used:

• Context from dashboard - CLAUDE.md (source)
• Context from dashboard - AGENTS.md (source)