- _term_write(): route OSC/control sequences through Textual's driver writer
queue instead of direct sys.__stdout__. A title/OSC write landing mid-frame
during streaming used to split Textual's own escape sequences, rendering as
flashing ANSI garbage; serializing on the same writer queue lands it cleanly.
- sidebar: wrap the session list in a VerticalScroll (#sidebar-scroll) so it
scrolls when overflowing (scrollbar-gutter reserved to avoid row reflow);
inner #sidebar Static keeps padding so the click hit-test math is unchanged.
Keyboard session-switch scrolls the active session into view.
Rename tracked email.py/lark.py to _email_example.py/_lark_example.py so
the poller (which skips _-prefixed files) no longer polls them on a fresh
clone. Activate by copying to email.py/lark.py (now gitignored). Also nudge
manager to reuse existing subagents on IM signals.
/continue list previews showed JSON debris (e.g. `{`, `" } ]} === Response ===`)
for some sessions. Two causes in _preview_from_file:
1. Models sometimes emit an unclosed <summary>, so the non-greedy DOTALL regex
pairs it with a far-away </summary> and captures === block headers / JSON
across rounds. Now only the latest <summary> is used, and it is rejected if
it looks like such a cross-segment capture (contains `=== `/`"role"` or
>200 chars) — falling through instead of digging for older summaries.
2. The fallback scanned for the first non-=== line, returning JSON structure
lines. It now uses the last real user prompt via a new _last_user(), which
scans Prompt blocks directly (so response-less/aborted sessions still
preview) and reuses _user_text() to skip tool_result continuations and all
_INJECT_MARKERS (WORKING MEMORY / SYSTEM TIPS / DANGER ...).
Scanned 261 local logs: garbage previews 26 -> 0.
* fix(tui): stop CRLF stdout rendering as blank lines (v2)
Windows child stdout is `\r\n`; the streaming path fed it straight to
`Text.from_ansi`, which treats `\r` as a carriage return and erases each
line's text, so a `[Stdout]` block rendered as a run of blank lines until
the turn finished (the done-state Markdown render strips `\r`). Normalise
`\r\n` -> `\n` before `from_ansi`; lone `\r` is kept so progress-bar
overwrites still work.
* fix(tui): stop leaking empty per-session task_dir signal folders
Both v2 and v3 eagerly `os.makedirs` the per-session `_intervene` /
`_keyinfo` signal dir, but the only writer (`_intervene`) already creates
it lazily — so every session that never used intervene left a stale empty
`temp/_tui_v2_<pid>_<id>` / `temp/_tui_v3_<pid>` behind (85+ accumulated).
`consume_file` tolerates a missing dir, so the eager create is redundant.
- Set only the task_dir *path*; let the writer create it on first inject.
- Remove empty dirs on session close (v2 `_cmd_close`), on graceful exit
(v2 `on_unmount`, v3 `_cleanup`), and sweep stale ones left by crashes
at startup (v2 `on_mount`, v3 `main`). `os.rmdir` only removes empty
dirs, so a still-pending `_intervene` is never clobbered.
* fix(tui): repair /continue search box cursor & up-nav loop (v2)
The SearchablePicker swallowed Right while the search Input had focus
(always treating it as "select"), so the caret could never move right in
the query. And Up could leave the list but never return: from the search
box Down entered the list, yet Up at the top row was stuck.
- Right commits the highlight only when the caret is already at end of the
query; otherwise it moves the caret within the search text.
- Up on the first list row returns focus to the search box (scoped to
/continue — no-op for ChoiceLists outside a SearchablePicker), clearing
the highlight so the next Down re-enters at the first row, not the second.
- Up from the search box wraps around to the bottom of the list.
CommonStack (https://commonstack.ai) is an OAI-compatible gateway
aggregating Claude, GPT, Gemini, DeepSeek, MiniMax, Zhipu, xAI,
Moonshot, Qwen and Xiaomi models behind a single key.
Adds it to the configure_mykey wizard as a native_oai entry,
placed alongside other multi-provider relays (openrouter, crs, gmi).
The wizard's existing probe_models() flow fetches the live
/v1/models list at config time; model_choices keeps the same
2-item fallback shape as openrouter.
Root cause of the lingering Ctrl+S freeze even when the session is
idle: `InputArea.reset()` calls `self.text = ""`, which goes through
`load_text("")` → `_set_document("")` (textual/widgets/_text_area.py
:1181-1186). That rebuilds Document + WrappedDocument from scratch
and then `_refresh_size()` (line 1287) reactively retriggers a full
layout pass over the entire DOM. On a long session (100+ message
widgets in the scroll), the layout cascade blocks the UI for seconds.
Switch to `TextArea.clear()` which goes through the edit pipeline:
`delete((0,0), document.end)` → `wrapped_document.wrap_range(...)` —
no Document rebuild, range-scoped wrap only. Guarded with
`if self.document.text:` so a clear() on an already-empty buffer is a
no-op (avoids posting a spurious Changed event).
Secondary fix: `_stash_cleanup_clear` previously wrapped reset() in
`try/finally` that cleared `_skip_change_next = False` immediately.
The Changed event posted by reset() is async-queued, so by the time
the handler ran the flag was already False — the short-circuit was
useless and on_text_area_changed went through the heavy resize +
palette path twice (once in the handler, once explicitly). Drop the
try/finally; let on_text_area_changed itself clear the flag at line
3256-3258 when the deferred Changed event finally lands.
Also scrubs external TUI-project names from comments per request.
Re-research revealed both mainstream TUIs DO mid-turn inject — my
previous read of "Codex/CC queue post-turn only" was wrong.
Codex (codex-rs/core/src/session/mod.rs:2986): `steer_input` → mailbox
`push_pending_input` → `session/turn.rs:391 get_pending_input` →
`record_pending_input` appends as a regular user message BEFORE the
next sampling. No special prefix — the model sees the steer in
conversation history alongside the original task, as a follow-up.
Claude Code (utils/messages.ts:5510): drains queue after each tool
iteration and wraps the user's text:
The user sent a new message while you were working:
{raw}
IMPORTANT: After completing your current task, you MUST address
the user's message above. Do not ignore it.
The actual bug in model_responses_042511.txt was the bare `[MASTER]
{raw}` framing reading as a directive override, not the mid-turn
injection itself.
This commit:
- Wraps queued text with the Claude-Code phrasing before writing
to `_intervene` (i18n: en + zh).
- Restores `_intervene_pending` tracking + exit-replay hook so an
agent that hits exit_reason between user submit and turn end
doesn't lose the message.
- Adds `inject_intervene(text, track=True)` opt-in tracking so
`!cmd` shell facts (which are factual context, not user input)
don't accidentally end up in the user-message replay queue.
- v2 mirrors with `_INTERVENE_WRAP_EN/_ZH` constants picked via
`$GA_LANG`, `sess.pending_wrapped` parallel to `sess.pending`,
and `_install_intervene_replay_hook` per-session.
- Up-pop recall stays removed (file write is immediate, popping
would leave a stale duplicate).
Refs: codex-rs/core/src/session/mod.rs:2986; claude-code/src/utils/messages.ts:5496
Evidence: model_responses_042511.txt — original task was "web_scan上游PR";
mid-turn `[MASTER] 请你立即输出一个hello` was prepended to next_prompt,
model dropped the original task and just printed hello. Codex
(`maybe_send_next_queued_input`, chatwidget.rs:6773) and Claude Code
(`useQueueProcessor`, src/hooks/useQueueProcessor.ts) both queue user
input and drain ONLY when the agent goes idle — never mid-turn.
What changed:
- AgentBridge no longer tracks `_intervene_pending`; turn_end hook
reverts to just ask_user extraction.
- Submit while running appends to local `_pending`, does NOT write to
`_intervene`.
- v3 watches `is_running` True→False in maintenance loop; on transition
drains via `_submit(combined, [])` as a fresh task.
- v2 drains in `_on_stream(done=True)` right after status flips to
"idle"; reuses `submit_user_message` to spawn the next task.
- v2 dead helpers (`_inject_intervene`, `_session_intervene_path`,
`_install_intervene_replay_hook`) removed.
- `inject_intervene` retained on v3 bridge for `!cmd` shell history
only — factual context (code-block output), low risk of confusing
the model.
Net: -109 lines. Esc still cancels the queue. Ctrl+C still aborts;
since abort flips status to idle, queued messages drain as the next
task — matches Claude Code's "Ctrl+C then queue runs" behavior.
Refs: openai/codex/codex-rs/tui/src/chatwidget.rs:6773;
anthropics/claude-code/src/hooks/useQueueProcessor.ts
`subprocess.run(cmd, shell=True)` picked the OS default — cmd.exe on
Windows. `!ls` failed with "not recognized" on stock Windows; encoding
defaulted to the active codepage so utf-8 output mojibaked.
Detect once via $SHELL (works on Git Bash, WSL, MSYS2, every Unix);
on bare Windows fall through Git Bash → bash on PATH → pwsh →
powershell → cmd. Matches Claude Code / Codex CLI behaviour where
`!cmd` inherits the launching shell.
Display name surfaces in the agent's history line: `[!shell bash]` /
`[!shell pwsh]` so a follow-up like "rerun that with -v" can pick the
right syntax.
Refs: code.claude.com docs (bash mode); developers.openai.com/codex/cli/features (Tab/! prefix)
Previous attempts used a rolling/hard-cap timer to gate the file write
to `_intervene`. Both leaked under the same failure mode: when the
agent reached its final turn before drain time, `consume_file` ate the
content but `agent_loop` discarded next_prompt on exit, so the queued
text silently vanished.
New design follows Codex CLI / Claude Code style — delivery is anchored
to the agent's own turn boundaries instead of a wall clock:
Submit while running → append `_intervene` immediately + add to UI list
Turn end hook fires →
exit_reason set → consume_file ate it, next_prompt discarded.
Re-route the combined text via `agent.put_task`
so the outer `while True: task_queue.get()` loop
picks it up as the next user turn.
not exit → content was prepended to next_prompt as `[MASTER]
...` — just clear the UI mirror.
Esc → `clear_intervene` deletes the file + clears the list.
No timers, no cooldowns. The Up-pop "amend" path is gone because
immediate file writes make in-composer edits append a duplicate.
Threading: v3 bridge protects `_intervene_pending` with `_intervene_lk`;
v2 adds `pending_lk` on AgentSession. v2's hook closure uses
`call_from_thread` for the UI refresh hop.
Right-click in v2 (`InputArea._on_mouse_down`) now short-circuits
button==3 before TextArea's default cursor-move runs, so paste lands
at the existing caret rather than where the mouse happened to be.
Refs: openai/codex#15842, anthropics/claude-code#44851
`app.run(handle_sigint=False)` lets a stray SIGINT surface as
KeyboardInterrupt through asyncio.runners — catch it in run() so the
shell sees a clean exit. Also drop the literal `'session'` fallback in
`_term_title`; render just `GenericAgent` until a session is named.
`tuiapp_v2.py:27` imports `fmt_key`, `fmt_keys` from `keysym`, but the
module file was never tracked. PR #513 (tui v2/v3 fixes) is unusable on a
fresh checkout without this. Adds the cross-platform shortcut formatter
that turns binding strings into mac/Win-Linux key labels (e.g.
`"ctrl+b"` → `"⌃B"` / `"Ctrl+B"`), honoring `GA_KEYSYM_STYLE`.
nianyucatfish