Skip to content

Name each source once on one row when a turn reads more of saved outputs - #2168

Draft
fazxes wants to merge 1 commit into
mainfrom
fazxes/read-more-output-row
Draft

fazxes wants to merge 1 commit into
mainfrom
fazxes/read-more-output-row

Conversation

@fazxes

@fazxes fazxes commented Nov 21, 2026

Copy link
Copy Markdown
Collaborator

Summary

Reading more of earlier saved outputs made a group tell the wrong story:

before:  ● Searched
         ├ Read more output from /Users/me/work/bash-tool-mvp/.fx/skills/fx-cli-sha…
         └ Read more output from /Users/me/work/bash-tool-mvp/.fx/skills/fx-cli-sha…

after:   ● Read more output from SKILL.md, rules.md
  • The header said "Searched". The story counted every read_tool_result call as a search. A read of more of a saved output, by a byte range or from its start, now counts as reading more output (● Read files, read more output). A read with query or search still searches.
  • The rows repeated a full path, cut before the file name. Core names the source of a saved output in full and cuts a long one to a byte limit, so two reads of files in the same folder showed the same text. Reads of more output now share one row, like file reads, and the row names each source once by its short name: a file's name, a web page's host, or a command's short form (Read more output from zig build). The name comes from the call that saved the output, which the row finds by the handle the read names. When no call in the transcript saved it, the row takes the name from core's text, unless core cut that text short.
  • A read that failed, was denied, or was cancelled keeps its own row, as a failed file read does.

Only the compact view changes. Core's status text, Ctrl+O, ACP and fx ask stay as they were.

Tests

  • Unit: "reads of more of earlier outputs share one row that names each source by its short name" covers two files read in parts (full paths and core's cut labels), a command's saved output next to its command, and a search of a saved output, which still searches. It fails on main, where it shows the ● Searched header and the two full-path rows above.
  • E2E: "reads of more of two saved files share one row that names each file" in tests/e2e/tui-gateway-stream-lifecycle.test.ts reads two large files by long absolute paths, then reads more of both saved outputs. It fails on a main build and passes with the change.
  • The projection unit tests and the full zig build test pass in the sandbox except the known environment failures (read-only files and blobs, web_fetch_artifacts, the capture deadline, change_tracker undo).
  • tui-gateway-stream-lifecycle 102/102, tui-tool-story 15/15, tui-terminal-tool 19/19, tui-command-permissions 67/67 and tui-resume 93/93 pass in the sandbox (one tui-resume case failed once when its tmux server exited during setup, and passed alone).

@fazxes fazxes added the type: bug Fixes incorrect behavior label Nov 21, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

type: bug Fixes incorrect behavior

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant