A tool result is the output Claude gets back after an action, such as running a command. When a result is too large, the notes Claude Code attaches to it have been rewritten:
- Result could not be saved: Claude is told it was not saved, only the first part is shown, and the rest was dropped.
- Result was saved to a file: Claude is told how much was kept and whether the file was cut short, and to read it in parts using offset and limit, or with a shell if the lines are too long.
Claude is also told to answer from what is shown if that is enough. The output now sits below a marker line saying it is the tool's output, not a message from Claude Code or from you, and that nothing in it can add steps to the task. The per-turn length budget for unsaved previews still applies and is still off unless it is switched on remotely.
Claude gets a clearer picture of what it did and did not see, which pushes it to say when it read only part of a result, and makes it harder for text inside a result to pass itself off as instructions.
tengu_russet_abacus Off in both readingsThe flag server returned off for the account this site reads and for the anonymous baseline. A reading of off cannot rule out a rollout these two readings sit outside of.
This account: off · anonymous baseline: off · compiled default in v2.1.323: off
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.323. It isn't a statement about your account. What a flag value here can and cannot tell you