Newer Claude models (Opus 4.8, Sonnet 5) invent extra fields in tool calls, breaking Pi's edit harness

Better Models: Worse Tools

Armin Ronacher found that Claude Opus 4.8 and Sonnet 5 sometimes add invented keys like requireUnique or oldText2 to Pi's edit tool calls, causing schema validation failures. Older models don't do this. In multi-turn agent sessions, Opus 4.8 fails roughly 20% of the time; stripping thinking blocks halves the rate, and strict tool invocation eliminates it. He suspects Anthropic's newer post-training is tuned for Claude Code's own flat edit tool, whose client silently absorbs malformed calls, so the model never gets penalized for inventing extra fields.

Why it matters: Armin Ronacher's hands-on test shows Opus 4.8 and Sonnet 5 hallucinate extra fields in Pi's edit tool schema ~20% of the time, while older models don't. It's a concrete, reproducible engineering finding with direct relevance for agent builders. Not scored higher because the is...