What
- Tool-result rendering now recognizes a
files_readresult and shows "saved N of M file(s)", optionally "under " when a shared save directory is used; the WebFetch-style summary can also show "read N file(s)" for multiple paths. - The artifacts tool's
read_fileaction now accepts apathsarray instead of only a single path, so several published files can be requested in one call and saved under a shared directory. - The result formatter reports how many of N files were saved, lists each with size/content-type/sha256 (or an error), inlines full text for files under a size budget, and truncates the listing when the combined result would be too large.
- Path resolution for
read_filewas updated to handle the multi-path case using a different helper than the single-file path. - Multi-path input is validated (duplicate paths are removed, a "too many paths" case is reported), but a batched read that needs permission is currently always denied, telling the model to read files one at a time first, since batch reads don't yet trigger their own permission prompt.
Why
This lets Claude read and save several files from an artifact in a single call instead of one at a time, though permission handling for batch reads hasn't caught up yet, so a batch request needing approval will currently fail until read singly.
Names in the bundleread_file
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Confirmed since
Anthropic's documentation has since written up read_file, on inference-hooks-endpoint.
"tool_name": "read_file",manage-claude/inference-hooks-endpoint see the edit
One source agreesOne thing we can check says the same as this entry.
Anthropic's documentation agrees
Anthropic's documentation has since written up read_file, on inference-hooks-endpoint.