O10 — Transcript and tool visualizations¶
Status: design stub. Back to the component map.
Purpose¶
Make messages, reasoning that the provider exposes, tool activity and results readable at several levels of detail. Reuse the same rendering pieces in live sessions, history and trace inspection.
Existing foundation¶
The existing history tools supply records and forensics. AIUI's current catalog includes JsonView, TextView, ToolLog and PromptRecordView; ToolLog's own application-tool stream is not automatically an agent transcript.
Proposed boundary¶
Keep native item IDs and original data behind derived views. Separate partial output from completed replacements; mark missing/truncated data and unavailable assets. A renderer must not execute tools or create model work when inspecting history.
Inputs and outputs: Transcript items and status/coverage metadata → readable timeline entries, expandable tool views and links to raw evidence.
Candidate progression¶
- Mock: Streaming text, parallel calls, a long command, a patch/diff, a rich result, an error and a missing-output gap.
- First usable slice: Render a real session and a recorded session with shared components; compare what the data actually supports.
- Later: Specialized renderers, linked source/files, reasoning/context inspection and configurable density without flattening unsupported native fields.
These are discussion candidates, not scheduled work or implemented capabilities.
Recorded review — 19 October 2026¶
The O02 comments ask for several levels of detail, from unobtrusive activity to raw JSON, useful to humans and debugging agents. Expanding detail should not require every session to fetch/parse all raw evidence continuously.
The O17 request adds meaningful file links: retain host/working-directory context, open a read-only source view, annotate or select code and carry the selection into a prompt. Historical evidence and current file contents must remain distinguishable.
Your additions¶
- What should be collapsed by default, and what must remain visible during execution?
- Which tool-result types deserve specialized presentation first?
Additional requirements, examples and decisions: TBD with you.