Raise your tool success rate
Failed tool calls waste a full round-trip each. Give the agent what it needs to get them right first time.
Your tool success rate is the share of tool calls — commands, edits, reads — that complete without an error.
Why it matters
Every failed tool call (a bad command, a wrong path, an interrupted run) costs a round-trip and some output tokens, and then needs another call to recover. A low success rate is a quiet tax on every session.
How to improve
- Give correct context up front. The right working directory, file paths, and any environment quirks (a non-standard build command, required env vars) prevent the most common failures.
- Let it read before it writes. An agent that inspects a file or runs
--helpfirst makes fewer wrong guesses. - Name the tools and constraints. If a command must be run a particular way, say so — don’t make the agent discover it by failing.
- Watch for interrupts. Hard-stopping a tool mid-run shows up as a failure; see reduce interruptions.