Collapse large pastes; remove stale queued-message notice - #7
Merged
Merged
Conversation
Pasting something long (e.g. a stack trace, file content, or diff) used to dump the entire text inline into the composer, making it unreadable and pushing the actual prompt out of view. - ChatComposer.on_paste: pastes over PASTE_COLLAPSE_THRESHOLD (100 chars) are collapsed to a "[LINES N]" placeholder instead of being inserted verbatim; shorter pastes are inserted as before. - The real pasted text is tracked internally and substituted back in when the message is submitted (Enter/Ctrl+S), so the model still receives the full original content - only the on-screen composer is compressed. - Same-line-count pastes within one draft get distinct placeholders (`[LINES N]`, `[LINES N#2]`, ...) so each expands back to its own text. - Added tests in tests/test_composer.py covering collapse, the threshold boundary, expansion on submit, and disambiguation of duplicate placeholders. Co-Authored-By: Warp <agent@warp.dev>
ChatPane now tracks each "📥 Queued (#N)" SystemMessage widget in self._queued_notices, parallel to self.state.message_queue. When a queued prompt is dequeued and starts its turn, the matching notice is removed from the chat log instead of staying there indefinitely. Notices for prompts discarded on cancellation are removed the same way. Co-Authored-By: Warp <agent@warp.dev>
- ui/tui.py: add Escape keybinding to request_cancel and map tool_start/tool_done/tool_error/tool_progress to readable trace labels. - main.py: break the agent tool loop on any tool execution error instead of letting the model retry blindly; keep plan-mode write rejection as a soft policy nudge so the model can recover with a real plan. - core/workspace_tools.py: tighten the generic tool-leak guard so natural language mentioning "arguments"/"path"/"content" near HTML-like "<" is not treated as a malformed tool call. - tests/test_workspace_tools.py: add regression test for the false-positive prose case. Co-Authored-By: Warp <agent@warp.dev>
Revert the break-on-tool_error behavior in main.py. Malformed tool calls and tool execution failures are now appended to the conversation history as context so the model can self-correct on the next turn. The user interrupts via Esc (and ctrl+c) only. Add regression tests verifying the loop continues after a malformed tool call and after a tool execution failure. Co-Authored-By: Warp <agent@warp.dev>
ui/tui.py: on_trace_event now builds short, human-readable sentences for tool_start/tool_done/tool_error instead of empty or key=value details. For example: - tool.start → "about to run read_file on `scraper/base.py`" - tool.done → "read_file finished on `scraper/base.py`" - tool.error → "read_file failed on `scraper/base.py`: file does not exist" main.py: include `path` in tool_done and tool_error trace payloads so the UI can show the affected file/path. tests/test_workspace_tools.py: regression tests already cover the continue-after-error behavior. Co-Authored-By: Warp <agent@warp.dev>
- ui/tui.py: cap conversation history passed to the model at the most recent 8 turns so long sessions cannot drown out the latest user message. - core/workspace_tools.py: add a system reminder to always answer the most recent user message, not earlier conversation turns. - core/providers.py: reduce httpx default timeout from 120s to 30s so a stuck provider call returns faster and Esc/cancel feels responsive. Co-Authored-By: Warp <agent@warp.dev>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Follow-up to #6 (already merged): two small composer/queueing UX fixes that landed on the same branch after that PR was merged, so they need their own PR.
Changes
ui/tui.py): pasting more than 100 characters into the chat composer now collapses to a[LINES N]placeholder instead of dumping the raw text inline. The real content is tracked internally and substituted back in on submit, so the model still receives the full text. Duplicate same-line-count pastes in one draft get distinct placeholders ([LINES N],[LINES N#2], ...).ui/tui.py): the "📥 Queued (#N) — will run once the current task finishes" notice now gets removed from the chat log once that queued message is dequeued and starts running (or discarded, with its notice removed, if the run is cancelled). Previously it stayed in the log indefinitely even after being consumed.Testing
Full test suite passes locally:
PYTHONPATH=. .venv/bin/pytest tests/ -q→ 70 passed.Co-Authored-By: Warp agent@warp.dev