feat: pin research subagents to DeepSeek Flash
This commit is contained in:
@@ -2,6 +2,15 @@
|
||||
|
||||
## Unreleased
|
||||
|
||||
- New `/agent [filter]` OpenCode-style chat inspector: a full TUI panel shows the
|
||||
agent tree on the left and the selected agent's live chat transcript on the
|
||||
right (`↑↓` select, `enter`/`tab` open chat, `t` toggle thinking, `o` open the
|
||||
full session, `esc` close). Transcripts are parsed from the agent session
|
||||
JSONL (user/assistant/tool-call/tool-result/thinking blocks) and refresh every
|
||||
1.5s while an agent is streaming; the tree shows status glyphs, elapsed time,
|
||||
and per-agent tasks. `/agent-open` keeps the previous session-switching flow
|
||||
(`Subagent Actions → Open`), with `/back` returning to the main session.
|
||||
|
||||
- New `/agent [filter]` command opens a subagent's full chat session in the TUI
|
||||
(OpenCode-style `Subagent Actions → Open`): a selector lists the run's agents
|
||||
with status glyphs, elapsed time, and task; picking one switches the session
|
||||
|
||||
Reference in New Issue
Block a user