The + menu lists up to three agents that have actually been run, beside
Local and SSH. Other Agents… opens Search Everywhere already filtered to
agent. New Agent Tab still starts the one used last.
Refs #955
Co-authored-by: stevelliu <stevelliu@tencent.com>
* feat(agents): add Qoder CN CLI integration
* fix(agents): spell Qoder CN's config override QODERCN_CONFIG_DIR
The China build of Qoder resolves its configuration through QODERCN_CONFIG_DIR.
The first pass invented QODER_CN_CONFIG_DIR instead, so an install that set the
real one was treated as unrelocated: hooks went to ~/.qoder-cn and history was
scanned from a directory the CLI never writes to. The test that should have
caught it set the same invented name, so it only proved the name agreed with
itself.
Picks up the rest of the review as well — detect the `qoder-cn` dispatcher
beside the other two binaries, call it the mainland-China build rather than a
different vendor, and drop the QODER_CN_HOOK_EVENTS alias for the table it only
pointed at. The serialized enums keep their variants appended; only the
independent ALL arrays moved QoderCLICn next to QoderCLI.
The Window & Tabs page is gone. Its three remaining tab settings (new tab
position, tab bar position, auto grouping) are a Tabs group on General,
below Startup & restore, so the nav has seven sections.
Removed outright:
- SSH tab title (`ssh_tab_title`, #726, unreleased). The sidebar already
groups SSH tabs under their host, and renaming a tab pins its name.
- Open diff preview from sidebar counts (`sidebar_diff_preview`, #247).
The sidebar counts and the Info panel's changes row always open the
diff overlay; the large-tree stall it worked around is bounded. An old
config.json that still carries either key loads as before.
The now-unused window settings icon goes with the page.
* feat(search): replace the command palette with tabbed Search Everywhere
The palette was one flat list that every new kind of row had to be squeezed
into: tabs and SSH hosts rode along as "Switch to Tab: …" and "SSH: …"
commands, and the only way to narrow to one kind was a magic seed word.
Search Everywhere splits it into sources behind one trait — All, Actions,
Terminals, Hosts — each with its own empty-query layout and ranking. The All
tab shows each source's top rows, ordered by best match, with a row that
opens the full tab. Tab / Shift-Tab walk the tabs and keep the query.
- Terminals lists every open tab of every workspace (reusing the switcher's
tab rows) and jumps to it wherever it lives, plus the shells and agents.
- Hosts replaces the separate "Add Connection" input: a typed address or
full `ssh …` line offers to connect.
- Fixes Return doing nothing after a search that found nothing, or when the
search opens pre-filtered: gpui-component re-picks the row from a stale
frame; the delegate now re-arms the first row.
- Fixes `ssh -p 2222 me@box` being offered as a quick-connect address with
user `ssh -p 2222 me`.
The keymap action stays `TogglePalette` so custom bindings keep working.
* feat(search): a Sessions tab to resume past agent sessions
Search Everywhere gains a Sessions tab listing the Claude Code and Codex
sessions on this computer, read from ~/.claude/projects and
~/.codex/sessions ($CODEX_HOME). Sessions that ran in the focused tab's
directory lead; the All tab offers the last three of them before anything
is typed. Return opens a new tab in the session's directory and runs the
agent's resume command with its configured launch flags.
- Only each transcript's head and tail are read, off the window thread,
and cached by path, size and mtime: ~170ms cold for 50 sessions,
under 1ms warm. The search opens on the cached list and fills in.
- Titles: /rename name, then the agent's own title (ai-title, Codex's
session_index.jsonl), then the first thing typed, skipping harness
injections. Codex rollouts it ran for itself (subagent/internal) and
sessions never asked anything are left out.
- A session whose directory is gone is refused with a notice rather than
resumed where the agent cannot find it.
* feat(agents): recognise Empryo and Prime Agent
Prime Agent (PrimeIntellect-ai/prime-agent) is a Pi fork: detected as
`prime-agent`, resumes with `--resume <id>`, forks with `--fork <id>`,
`--no-session` opts out, and reports status through the shared Pi
extension bridge at ~/.prime/agent/extensions/tty7/index.ts.
Empryo is detected as `empryo` and resumes with `empryo --session <id>`.
No hook installer yet: Empryo filters TTY7* variables out of hook
processes, so `tty7-app agent-hook` would stop at the missing marker.
* feat(agents): Antigravity status hooks
Antigravity CLI (`agy`) now installs a `tty7` hook set in
~/.gemini/config/hooks.json, next to the user's own sets:
PreInvocation -> prompt-submit, PostToolUse -> tool-complete,
Stop -> stop. Only the first PreInvocation of a turn (invocationNum 0)
counts as a new prompt, later ones keep the turn working; a Stop with
fullyIdle false is ignored. conversationId and workspacePaths feed the
session id and cwd. PreToolUse is left alone because it must answer
with a decision.
* fix(agents): index Prime Agent and Antigravity in settings, scope conversationId alias
- Add settings titles, search keywords (en/ja/zh) and Agents index entries
for the two new HookAgents; agent_rows_are_in_the_search_index requires
one per HookAgent::ALL.
- Read Antigravity's conversationId as the session id only for antigravity:
the alias table is last-write-wins, so a global alias could replace
another agent's session id.
- Drop the stray .empryo/ job records.
---------
Co-authored-by: kalpak <you@example.com>
Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
* feat(agents): quick launch for detected CLI agents (#955)
Every agent whose launch program is on PATH becomes a palette command,
"Agent: <name>", ordered by frecency and bindable as LaunchAgent:<slug>.
"New Agent Tab" (Cmd+Shift+A on macOS; unbound elsewhere, where
Ctrl+Shift+A is select-all) launches the most recently used one, and the
New Tab menu gains a single "Launch Agent..." row that opens the palette
pre-filtered to them.
A launch always opens a new pane (a tab in the active tab's cwd, or a
split when picked from the palette with Alt held) and types the command
into that pane's shell once it exists - immediately for a local pane, on
landing for a remote one via PendingSpawn::run_on_land - never into a
pane that was already there. Detection, status and resume then work as
for a hand-typed agent.
The command is the agent's bare binary unless the new `agent_launch`
config map overrides it. A wrapper named there is detected as its agent
without an `agent_commands` entry: the daemon folds the programs
`agent_launch` runs into its alias map (interpreters, shells and real
agent names excepted), reloaded when config.json changes instead of
once per process, and a mapped script run under its interpreter
(`bash ~/bin/cc`, `node cc.js`) is now recognised too.
A running agent's pane menu gets "Set Current Launch Args as Default",
which writes its launch argv - minus session flags and positional
prompts, joined with the settings' quoting - into `agent_launch`.
Remote workspaces cannot be asked for their PATH through the Host
trait, so they offer the agents previously seen running in that
workspace (WindowView::seen_agents).
* fix(agents): count only agents that start under a view, not ones reattached to
A view rebuilt over a running pane - every agent tab after an app
restart, or a workspace switched back to - saw its agent appear from
nothing and reported it as detected, bumping that agent's frecency once
per launch of the app. The terminal now remembers whether its link was
an attach, and such a view only reports agents once its shell has been
seen back at the prompt with no agent in front.
Also pins down that the agents seen in a remote workspace, its quick
launch list, persist in views.json with the rest of the workspace.
* fix(daemon): probe the foreground as soon as a new program takes it
The foreground probes ran on output only, at most once per 500ms
interval. A quick launch types the agent's command the moment the shell
is up, so the agent drew its whole first screen inside the interval of
the prompt it was typed at and then waited for a key: the pane never
learned it was running an agent until something else printed. Seen in
a dev instance, where a launched Claude Code stayed undetected at its
trust prompt.
The reader now asks the pty for its foreground process group on every
read (one ioctl) and probes immediately when it changes.
Every placeholder frame in docs/ now shows a real capture of the current
build: 20 screenshots plus two short looping clips (prompt editor, pane
drag). The README and the docs home page link a one-minute tour covering
agent status across repos, one agent driving another through the CLI,
the prompt editor, diffs, pane dragging, and sessions surviving a quit.
The interactive cursor-agent TUI fires beforeSubmitPrompt and stop; only
print mode (-p) lacks them. Mapping postToolUse to prompt-submit froze the
activity counter (opportunistic git refresh, agents --changed), shortened
turn timing and hid tool-less turns. Map it to tool-complete instead.
Cursor's ~/.cursor/hooks.json is a flat hook map — `command` directly on
each entry under `hooks.<event>` — so it reuses the flat writer Crush
introduced. Two things are Cursor-specific:
- The file needs `"version": 1` at its root or Cursor ignores it. tty7
now writes it when it creates the file or finds it missing, never
overwrites a version the user set, and reports a versionless file that
holds our hooks as Outdated so refresh repairs it.
- Payloads name the session `conversation_id` (only sessionStart repeats
it as `session_id`), and most events carry `workspace_roots` instead of
`cwd`. Both are now read as aliases, which is what Copy Session ID and
resume were missing.
Events: sessionStart, beforeSubmitPrompt, postToolUse, stop, sessionEnd.
cursor-agent does not fire beforeSubmitPrompt today (a known gap on
Cursor's side), so postToolUse also reports prompt-submit: the first tool
call opens the turn and Cursor's stop, which the CLI does fire, closes it.
A turn with no tool calls never shows as working. None of Cursor's
permission hooks are installed, since those would block on the empty
stdout tty7's hook prints.
CodeBuddy Code takes Claude Code's hooks as-is: the same nested
`hooks.<Event>[].hooks[{type, command}]` shape in ~/.codebuddy/settings.json
(or $CODEBUDDY_CONFIG_DIR/settings.json), the same event names, and the same
session_id/cwd payload fields. So it rides the shared hook-map installer and
needs no payload aliases.
The event table follows Qoder rather than Claude: CodeBuddy has a
first-class PermissionRequest and Elicitation, so it gets no Notification
hook, and StopFailure ends a turn like Stop. CodeBuddy also emits
SessionStart with source "compact" mid-turn, which is filtered out the same
way Qoder's is so the pane does not drop back to idle.
Detection covers all three npm bins (codebuddy, codebuddy-code, cbc).
Resume is `codebuddy --resume <id>` and fork appends --fork-session; stale
session and worktree flags are dropped from the replayed launch argv, and
--no-session-persistence disables both commands.
Icon, en/ja/zh names and search keywords, and docs are updated.
Crush fires only PreToolUse, and it mapped to prompt-submit. With no Stop
behind it the pane stayed on Working until Crush exited: every close asked
whether to cut Crush's work short, the tray and dot stayed on working, and
fork warned mid-turn. Map it to tool-complete, which still records
session_id/cwd for resume and bumps activity without moving the status.
* feat(agents): wire hooks, resume and detection for Kimi Code
Kimi Code CLI takes its hooks as [[hooks]] entries in the same
config.toml that holds the user's providers and models, so this adds a
third install strategy — a format-preserving TOML merge on toml_edit —
beside the JSON map merge and the owned files. Like Qwen it reports
permission requests first-class, so it gets no Notification hook.
Resume rides `kimi --session <id>`; fork stays unwired, Kimi
documents none.
Closes#693
Signed-off-by: Austin Spraggins <spragginsdesigns@gmail.com>
* fix(agents): harden the Kimi Code TOML hook merge and its resume flags
The TOML merge strategy the Kimi wiring introduces round-trips a shared
config.toml cleanly, but three gaps sat behind it.
`hooks_state` counted only the marked entries that still named an event,
so a hand-edit that dropped the key off one of nine entries left the
remaining eight matching the roster exactly and the file reported
Installed with a broken entry in it. Every marked entry now counts,
which is what the JSON merge already did and what `refresh_hooks` needs
to see.
A `hooks = []` spelled as an empty inline array made install fail
outright -- toml_edit keeps an empty array and an array of tables apart,
but the two say the same thing and neither carries any configuration. It
is now promoted rather than refused. Every other wrong-shaped `hooks`
key -- a string, a table, a non-empty inline array -- still refuses with
the file left byte-for-byte alone.
`Stop` is not the only way a Kimi turn ends: its own event reference says
`Stop` does not fire on interrupts and `Interrupt` fires instead, and a
turn that dies on an error reports `StopFailure`. Without those two an
Esc or a failed turn left the pane on "working" for good and `tty7 wait`
could only ever time out. Both are observation-only events and report
the same end of turn `Stop` does.
On resume, `--agent` and `--agent-file` join the stale flags: Kimi
rejects either next to `--session` at startup, and resuming rebinds the
session agent by itself, so replaying them turned a working resume into
a launch error.
Tests cover the wrong-shaped `hooks` keys, a config.toml that does not
parse on both install and uninstall, a file that does not exist yet, a
second install being byte-for-byte the first, mangled and surplus marked
entries, an uninstall threading between the user's own entries and the
tables after them, and the `--session=<id>`, bare `--session`,
`--continue` and `--agent` spellings on the resume path.
---------
Signed-off-by: Austin Spraggins <spragginsdesigns@gmail.com>
Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
* feat(cli): wait on commands, press keys, and reap orphan panes
`tty7 wait` was the orchestration primitive for agents only. A pane with
nothing reporting agent status read as `idle`, so `--until idle` returned
success instantly about a shell that was midway through a build, and there
was no state that meant "the command finished" at all.
Panes with no agent now report `no-agent`, and `free` ends the wait when the
foreground command has exited — the process-tree question `procs` could
already answer but nothing could block on. `send --key` covers the keystrokes
text cannot express, which is what a worker stopped at `waiting` is usually
asking for. `pane close` takes several panes and `--orphans` clears what an
interrupted `run` leaves behind. `doctor` finally performs the hooks check
its own help has advertised.
The skill shipped in this repo predated `wait` entirely and taught a
hand-rolled `procs` polling loop with no notion of delegation; it now covers
the loop, and its agent statuses, `ws rm` orphan claim and not-implemented
list are corrected against the code.
* fix(cli): close the gaps review found in wait, --key and pane close
Five things the first pass got wrong, in the order they bite.
`--until free --changed` waited on a command it had already missed: the
"something ran" edge is only set by a poll that catches the pane busy, and a
command that starts and finishes inside one 500ms interval never is. That is
indistinguishable from a command that never ran, so the timeout now names both
doors instead of letting a finished build read as a hang.
`free` also outranked the agent ladder, which is backwards. A pane whose depth-0
process *is* the agent — the tree cannot tell that apart from a shell at its
prompt — reads free for its whole turn, so a `waiting` the caller explicitly
asked for could be overwritten by a process-tree fact and then withheld by the
`--changed` rule that comes with it. `free` is now consulted only when none of
the requested agent states answered, which is both cheaper and what the docs
already claimed. An empty process tree is "we could not see in" rather than
"free" for the same reason `no-agent` exists.
`--key M-X` sent `ESC x`: the whole spelling was folded to lowercase, which is
free for Ctrl (the C0 rule clears the case anyway) and wrong for Alt, where the
character rides through as itself.
`send --help` listed the key vocabulary by hand next to the table it is a list
of; it had already drifted by one alias. It is generated now.
And a `pane close` batch that could not close everything raised an error, which
left `--json` holding prose exactly when a cleanup script needs to know which
panes are still its problem. It exits 1 with `{"closed":[…],"failed":[…]}`, with
the complaint still on stderr so `-q` reports it.
---------
Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
Audited every page under docs/ against the source. Fixes for what the
code actually does:
- agents: the status vocabulary is idle/working/waiting/done, not
running/waiting/idle; hook rows grow a separate Uninstall button; the
Settings table labels read "Copilot CLI" and "Grok Build"; Copy Session
ID lives in the tab's context menu, not the pane's
- cli: `pane ls --all` reports the owning workspace id, not "tty7-cli";
document bare `tty7 [PATH]` as the GUI launcher it is instead of listing
it as unimplemented; note `active_tab` and the `diagnostics` array; wait
also defaults to $TTY7_PANE
- git: the branch dropdown is a plain list with no search box and no
stash-and-switch, and checkout is not a palette command; quote the diff
overlay's own overflow notice rather than the sidebar's
- window: the unread marker tracks a finished agent turn, not any output;
rows cannot be dragged across groups; the sidebar and `tty7 tab ls`
resolve labels differently; drop Toggle Commit History and Checkout to
from the palette's Git group; ~/.ssh/config aliases are not palette
entries
- terminal: Ctrl+R dedups by command text and shows no directory; Esc does
not dismiss a ghost suggestion; document Cmd+Enter
- remote: GSSAPI is an ordinary Auth choice, not a managed-connection-only
mechanism
- fonts: Maple Mono NF CN leads the chain on Windows and Linux only; list
the real per-platform defaults
- settings paths: the three Links settings and per-pane history were filed
under the wrong sections
The in-app switch that wrote `~/.claude/skills/tty7-orchestration` was
removed in da6df70, before any release carried it — `git grep` finds it in
no tag. The changelog entry recording it landed the same day, written from
the branch's state rather than the tree's, and the documentation site (#478)
was then written from the changelog, so the feature reappeared on three
pages describing something the app has never shipped.
The orchestration page now says what is true: nothing is installed for you,
the agent-facing contract is `skills/tty7` and you add it yourself with
`npx skills add l0ng-ai/tty7`. That skill covers driving panes; `wait` is
documented on the orchestration page itself, so neither page promises what
the other holds.
38 pages under docs/, written against the source rather than the README:
config keys and their clamps from core::config, default keybindings from
ui::keymap, every CLI verb and flag from tty7-cli, agent aliases and
hook/fork/resume support from core::cli_agent, and Settings paths taken
from the actual en-US strings.
docs/features.md and its zh-CN translation are retired — everything in
them now lives in a page of its own, plus the two things they carried
that nothing else did (IME input, the performance notes). README and
README.zh-CN point at docs/ instead.
Screenshots and videos are placeholders for now: docs/images/placeholder.svg
with a caption naming what each shot should be.
Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
LiuSirandstevelliu
ARNO
l0ng-ai
l0ng-ai
Fabrice Aneche
ayamir
Austin Spragginsandl0ng-ai
l0ng-aiandl0ng-ai