- npm/codewhale/README.md: remove DeepSeek-first language
- docs/INSTALL.md: scoop install codewhale (not deepseek-tui)
- Wire decision card overlay rendering in main render loop
- Decision cards now appear centered on transcript when active
Before the turn breaks at the thinking-only checkpoint, drain
any sub-agent completions that arrived between the last hold
check and now. If a child finished while we were running the
final status check, surface its sentinel immediately rather
than delaying it to the next turn.
- Thinking cells now qualify as detail targets alongside tools and sub-agents
- Space key on empty composer toggles the focused cell's collapsed state
- Status message confirms expand/collapse action
- Builds on existing collapsed_cells HashSet from mouse context menu
Implements structured choice cards for the v0.8.43 truth-surface tracker.
When Brother Whale needs user input, it surfaces a bordered card with:
- A question prompt
- Numbered options with (default) marker
- Arrow/j/k navigation and 1-9 number-key shortcuts
- Enter to confirm, Esc to cancel
- Decision results surfaced as status messages
The widget compiles and the keyboard routing is wired into the main
event loop. Rendering overlay wire-up follows in the next commit.
When a turn has been running for > 30 seconds, the footer now shows
a classified stall reason suffix:
- 'waiting for model' — API streaming in progress
- 'tools executing' — active tool calls running
- 'sub-agents working' — child sub-agents in flight
- 'compacting context' — context compaction active
- 'background jobs running' — shell tasks executing
- 'waiting — no recent activity' — turn stalled with no classified work
- Show fuller turn ID prefix (16 chars) for disambiguation in task_read/task_cancel
- Replace ambiguous 'X active (Y running)' with clear per-status breakdown
- Add y/Y yank affordances for copying turn ID and full status from Tasks panel
- Add yank hint text in Tasks panel footer
The previous resolved allocative 0.3.6 which
pulls hashbrown 0.16, conflicting with starlark_map's hashbrown 0.14
dependency. Restore the original lockfile and update only workspace
crates to 0.8.43 via .
- Separate model auto-routing from Plan/Agent/YOLO TUI modes across all
READMEs and docs/MODES.md
- Introduce Fin as the fast thinking-off deepseek-v4-flash seam for
routing, summaries, RLM child calls, and coordination work
- Document /goal as current session tracking (not a TUI mode), with note
that a future Goal work surface should stay distinct from --model auto
- Extend deprecation shim timeline from 'one release cycle' to
through v0.8.x in REBRAND.md and npm READMEs
- Fix selection_to_text test to expect inline thinking preview
(short completed thinking now renders inline without Ctrl+O
affordance)
DeepSeek native accepts reasoning_effort: high | max, with low/medium
compat-mapping to high and xhigh mapping to max. OpenRouter and Novita
use the OpenAI-standard unified scale: none | minimal | low | medium |
high | xhigh.
Mapping (CodeWhale internal → wire):
off → DeepSeek: thinking disabled, OpenRouter: thinking disabled
low → DeepSeek: high, OpenRouter: low
medium → DeepSeek: high, OpenRouter: medium
high → DeepSeek: high, OpenRouter: high
max → DeepSeek: max, OpenRouter: xhigh
Previously the low/medium/high arm sent 'high' for all providers,
and the max arm sent 'max' for all providers — both wrong for
OpenRouter. Split each arm to preserve provider-native semantics.
Sources:
- https://api-docs.deepseek.com/guides/thinking_mode
- https://openrouter.ai/docs/guides/best-practices/reasoning-tokens
OpenRouter and Novita use OpenAI-standard reasoning_effort values
(xhigh|high|medium|low|minimal|none), not DeepSeek-native (max|high).
The xhigh/max/highest arm previously sent 'max' for all providers,
causing HTTP 400 on OpenRouter. Split OpenRouter+Novita into their
own arm that sends 'xhigh' instead.
Account for the help modal border and padding when calculating the visible
row window, so the selected command is not clipped after scrolling.
Also wrap Up/Down navigation at the list edges and strengthen the selected
row highlight for clearer focus.
- New github_close_pr tool distinct from github_close_issue; proper PR wording
in tool output, audit records, and gh pr close (not issue close)
- handle_read detects art_/call_/SHA refs and points to retrieve_tool_result
with copy-pasteable hints; error messages show correct tool for each ref type
- Shell delta tool results include the command field so the UI can resolve
task_id-only exec cells when the completion metadata arrives
- Sidebar background shell tasks show the actual command on the primary row
instead of just the task ID; task ID stays available as dim detail
- Tool routing falls back to task_id when exec_shell_wait has no command,
then updates when the completion carries command metadata
- Plan mode prompt explains update_plan as the handoff signal; model waits
for user action instead of continuing to tool around
- Base prompt clarifies handle_read scope (var_handles only) vs
retrieve_tool_result (artifacts/tool-result refs)
- New tests: close_pr_schema, close distinction wording, handle_read artifact
detection, shell_wait task_id fallback, sidebar background task labels
Hunter Bown