Compare commits

...
Author SHA1 Message Date
Miguel Palhas f02510465c docs(maestro): bare maestro opens the TUI 2026-08-25 12:34:36 +01:00
Miguel Palhas 8e251bf8be feat(maestro): default a missing subcommand to the TUI 2026-08-25 12:34:36 +01:00
Miguel Palhas 7cb14c7aae chore: empty commit to test the nix-autodeploy webhook
ci / web (push) Successful in 36s
ci / rust (push) Successful in 4m39s
ci / flake (push) Successful in 7m9s
2026-08-25 10:37:16 +01:00
Miguel Palhas 80a9399905 feat(maestro): session cards in the sidebar
ci / web (push) Successful in 29s
ci / rust (push) Successful in 4m25s
ci / flake (push) Successful in 6m34s
Turns each sidebar row into a card: name plus actions on line 1, an
adaptive meta line below carrying a colour-coded activity dot, the time
held in that state, any agent-reported detail, and the blocked-work or
exited markers that used to crowd the name row.

Renders the activity dimension the daemon already inferred and adds an
Error state fed by a Claude Code hook, which maestro installs globally
and addresses per session through MAESTRO_SESSION_ID / MAESTRO_URL.
Both clients implement one contract, docs/SIDEBAR.md.

Closes #101, #102, #103, #104, #105, #106, #107, #108, #110, #111, #112,
#113, #114, #115.
2026-08-24 23:26:59 +01:00
Miguel Palhas 9d9a2e723d merge: sidebar-cards/115-fourth-flake 2026-08-24 23:13:38 +01:00
Miguel Palhas bbafff5a98 fix(maestro): add settle-finished observable for natural-exit test
The no-retained-store test asserts a negative — the tree survives — so
waiting on that negative proves nothing: settle could still remove the
tree one instant after the assertion passes. mark_session_dead now flips
a per-record settled flag once the exit rule has been applied or refused,
exposed as SessionManager::is_settled; the test waits on it before
asserting.

Refs #115
2026-08-24 23:12:46 +01:00
Miguel Palhas 17e1dc4049 merge: sidebar-cards/114-hook-error 2026-08-24 23:01:14 +01:00
Miguel Palhas 436cb194c3 merge: sidebar-cards/113-more-flakes 2026-08-24 23:00:54 +01:00
Miguel Palhas 6471da2624 feat(maestro): report StopFailure as Error with the error spelling as detail 2026-08-24 22:54:21 +01:00
Miguel Palhas 2a43c76176 feat(maestro): report Active while a long tool call is still running 2026-08-24 22:52:51 +01:00
Miguel Palhas 54d84e3eaa docs(maestro): settle Error report TTL and clearing rulings 2026-08-24 22:46:46 +01:00
8 changed files with 255 additions and 61 deletions
+26
View File
@@ -62,6 +62,11 @@ pub struct SessionRecord {
/// resumes and [`Session::activity_detail`] is cleared with it. This
/// field is daemon-local bookkeeping only — it never rides the wire.
pub reported_activity: Option<crate::activity::ReportedActivity>,
/// Flipped once the natural-exit settle step (#82) has been applied or
/// refused for this session. `Dead` is observable before settling ends
/// (#115), so this is the observable that says its outcomes are safe to
/// assert on. Daemon-local bookkeeping; never rides the wire.
pub settled: bool,
}
#[derive(Debug, thiserror::Error)]
@@ -294,6 +299,17 @@ impl SessionManager {
sessions.get(&id).map(|r| r.session.status)
}
/// Whether the natural-exit settle step (#82) has finished for this
/// session: `Some(true)` once applied or refused, `Some(false)` while
/// still pending, `None` if the session is unknown. `Dead` is observable
/// before settling ends (#115), so anything asserting on a dead
/// session's worktree waits on this rather than inferring settle
/// completion from its side effects.
pub async fn is_settled(&self, id: Uuid) -> Option<bool> {
let sessions = self.inner.sessions.lock().await;
sessions.get(&id).map(|r| r.settled)
}
/// Spawn a new session. On success returns the freshly-created `Session`.
pub async fn create(&self, req: CreateSessionRequest) -> Result<Session, SessionError> {
let (effective_cwd, worktree_path, worktree_base_repo) = match &req.worktree {
@@ -429,6 +445,7 @@ impl SessionManager {
worktree_branch: req.worktree.as_ref().map(|w| w.branch.clone()),
bridge,
reported_activity: None,
settled: false,
},
);
}
@@ -657,6 +674,7 @@ impl SessionManager {
worktree_branch: None,
bridge,
reported_activity: None,
settled: false,
},
);
let _ = self
@@ -1069,6 +1087,14 @@ async fn mark_session_dead(inner: &Arc<Inner>, id: Uuid) {
}
}
}
// Settle has now been applied or refused either way (#115): from here on
// its outcomes are safe to observe.
{
let mut sessions = inner.sessions.lock().await;
if let Some(record) = sessions.get_mut(&id) {
record.settled = true;
}
}
let _ = inner.sidecar.write(&session);
let _ = inner.events.send(CatalogEvent::SessionUpdated { session });
@@ -946,6 +946,14 @@ async fn natural_exit_without_a_retained_store_destroys_nothing() {
})
.await;
// The settle step has been applied or refused either way — here refused,
// since there is no store to record the tree into. Only from this point
// is "the tree survived" a settled fact rather than a race (#115).
wait_for_async(Duration::from_secs(3), || async {
manager.is_settled(s.id).await == Some(true)
})
.await;
assert!(
wt_path.exists(),
"nothing may be destroyed when it cannot be recorded"
+28 -6
View File
@@ -25,13 +25,25 @@ pub const HOOK_MARKER: &str = "maestro-claude-hook";
pub const SCRIPT_FILE_NAME: &str = "maestro-claude-hook.sh";
/// Claude Code hook event → `ActivityState` spelling posted to the daemon.
pub const HOOK_EVENTS: [(&str, &str); 3] = [
// Tool finished: the agent is demonstrably producing work.
///
/// `PreToolUse` is what makes a session `Active` *during* a long tool call —
/// `PostToolUse` alone only fires once it ends (#114). `StopFailure` is
/// Claude Code's own name for "the turn ended on an API error"; there is no
/// failure signal in `Stop` or `SessionEnd` payloads, so those never report
/// `Error`.
pub const HOOK_EVENTS: [(&str, &str); 5] = [
// Tool starting: the agent is demonstrably producing work, even if the
// tool itself runs for minutes.
("PreToolUse", "Active"),
// Tool finished successfully.
("PostToolUse", "Active"),
// Permission / idle notices: Claude is blocked on the user.
("Notification", "AwaitingInput"),
// Turn ended: waiting on the user again.
("Stop", "AwaitingInput"),
// Turn ended on an API error; the payload's `error` field is forwarded
// as the report's `detail` (#114).
("StopFailure", "Error"),
];
/// The shipped hook script. POSIX sh so any `/bin/sh` works.
@@ -39,17 +51,27 @@ pub const HOOK_SCRIPT: &str = r#"#!/bin/sh
# Installed by maestro (#106): report Claude Code activity to its daemon.
# Exits silently unless running inside a session maestro spawned (#103 env).
# Event -> ActivityState mapping:
# PostToolUse -> Active (tool finished; agent is working)
# PreToolUse -> Active (tool started; agent is working)
# PostToolUse -> Active (tool finished)
# Notification -> AwaitingInput (permission / idle notice)
# Stop -> AwaitingInput (turn ended; waiting on the user)
# StopFailure -> Error (turn ended on an API error)
state="$1"
[ -n "$MAESTRO_SESSION_ID" ] || exit 0
[ -n "$MAESTRO_URL" ] || exit 0
# Drain the hook payload so the parent never sees a broken pipe.
cat >/dev/null 2>&1
payload=$(cat 2>/dev/null) || true
body="{\"state\":\"$state\"}"
# StopFailure payloads carry the outcome in "error". Only a plain enum
# spelling is reported as the detail — anything else (quotes, backslashes,
# prose) would risk breaking the JSON body, and the spelling already names
# the failure.
detail=$(printf '%s' "$payload" | sed -n 's/.*"error"[[:space:]]*:[[:space:]]*"\([a-z_]*\)".*/\1/p')
if [ -n "$detail" ]; then
body="{\"state\":\"$state\",\"detail\":\"$detail\"}"
fi
curl -fsS --connect-timeout 1 --max-time 2 \
-H 'Content-Type: application/json' \
-d "{\"state\":\"$state\"}" \
-d "$body" \
"$MAESTRO_URL/api/sessions/$MAESTRO_SESSION_ID/activity" \
>/dev/null 2>&1 || true
exit 0
+112 -40
View File
@@ -7,7 +7,7 @@ use std::process::{Command, Stdio};
use super::*;
fn sh(script: &str, args: &[&str], envs: &[(&str, &str)]) -> std::process::Output {
fn sh(script: &str, args: &[&str], envs: &[(&str, &str)], payload: &[u8]) -> std::process::Output {
let mut cmd = Command::new("/bin/sh");
// `sh -c CMD NAME ARGS…`: NAME fills $0, so the real arguments start at $1.
let mut argv = vec!["-c", script, HOOK_MARKER];
@@ -26,21 +26,48 @@ fn sh(script: &str, args: &[&str], envs: &[(&str, &str)]) -> std::process::Outpu
.spawn()
.expect("spawn /bin/sh");
// Feed a Claude-Code-style payload on stdin; the script must drain it.
child
.stdin
.take()
.unwrap()
.write_all(br#"{"session_id":"x"}"#)
.ok();
child.stdin.take().unwrap().write_all(payload).ok();
child.wait_with_output().expect("wait")
}
const SESSION: &str = "01234567-89ab-cdef-0123-456789abcdef";
/// Accept exactly one HTTP request on a throwaway listener, answer 204, and
/// hand back the raw request text. Runs `run` with the base URL once the
/// acceptor thread is in place.
fn capture_request(run: impl FnOnce(String) -> std::process::Output) -> String {
let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
let addr = listener.local_addr().unwrap();
let handle = std::thread::spawn(move || {
let (mut sock, _) = listener.accept().unwrap();
let mut buf = [0u8; 4096];
let n = sock.read(&mut buf).unwrap();
sock.write_all(b"HTTP/1.1 204 No Content\r\nContent-Length: 0\r\n\r\n")
.unwrap();
String::from_utf8_lossy(&buf[..n]).to_string()
});
let out = run(format!("http://{addr}"));
assert!(out.status.success(), "{:?}", out.stderr);
handle.join().unwrap()
}
fn session_envs(url: &str) -> Vec<(&'static str, String)> {
vec![
("MAESTRO_URL", url.to_string()),
("MAESTRO_SESSION_ID", SESSION.to_string()),
]
}
fn owned_envs<'a>(envs: &'a [(&'static str, String)]) -> Vec<(&'static str, &'a str)> {
envs.iter().map(|(k, v)| (*k, v.as_str())).collect()
}
/// The hook must be invisible outside maestro: no env vars (#103 unset) ⇒
/// exit 0, no output, stdin drained.
#[test]
fn script_exits_silently_without_maestro_env() {
for (event, _) in HOOK_EVENTS {
let out = sh(HOOK_SCRIPT, &[event], &[]);
let out = sh(HOOK_SCRIPT, &[event], &[], br#"{"session_id":"x"}"#);
assert!(out.status.success(), "{event:?}: {:?}", out.stderr);
assert!(out.stdout.is_empty(), "{event:?}: {:?}", out.stdout);
assert!(out.stderr.is_empty(), "{event:?}: {:?}", out.stderr);
@@ -51,46 +78,82 @@ fn script_exits_silently_without_maestro_env() {
/// activity endpoint and still exits 0.
#[test]
fn script_posts_mapped_state_to_the_activity_endpoint() {
let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
let addr = listener.local_addr().unwrap();
let server = move || {
let (mut sock, _) = listener.accept().unwrap();
let mut buf = [0u8; 4096];
let n = sock.read(&mut buf).unwrap();
let raw = String::from_utf8_lossy(&buf[..n]).to_string();
sock.write_all(b"HTTP/1.1 204 No Content\r\nContent-Length: 0\r\n\r\n")
.unwrap();
raw
};
let handle = std::thread::spawn(server);
let url = format!("http://{addr}");
let session = "01234567-89ab-cdef-0123-456789abcdef";
// The installer bakes the mapped state into the command's argv; the
// script relays whatever it is given.
let out = sh(
HOOK_SCRIPT,
&["Active"],
&[("MAESTRO_URL", &url), ("MAESTRO_SESSION_ID", session)],
);
assert!(out.status.success());
let request = handle.join().unwrap();
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Active"],
&owned_envs(&session_envs(&url)),
br#"{"session_id":"x"}"#,
)
});
assert!(request.starts_with("POST "), "{request}");
assert!(
request.contains(&format!("/api/sessions/{session}/activity")),
request.contains(&format!("/api/sessions/{SESSION}/activity")),
"{request}"
);
assert!(request.contains(r#""state":"Active""#), "{request}");
}
/// Every event maps to the state the issue specifies.
/// Every event maps to the state the contract specifies.
#[test]
fn event_state_mapping_matches_the_contract() {
let map: std::collections::HashMap<&str, &str> = HOOK_EVENTS.into_iter().collect();
assert_eq!(map["PreToolUse"], "Active");
assert_eq!(map["PostToolUse"], "Active");
assert_eq!(map["Stop"], "AwaitingInput");
assert_eq!(map["Notification"], "AwaitingInput");
assert_eq!(map["Stop"], "AwaitingInput");
assert_eq!(map["StopFailure"], "Error");
}
/// A `StopFailure` payload names the API error that ended the turn; the
/// script reports its `error` spelling as the report's `detail` so a card
/// can show why (#114).
#[test]
fn script_reports_the_error_spelling_as_detail() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Error"],
&owned_envs(&session_envs(&url)),
br#"{"session_id":"abc","hook_event_name":"StopFailure","error":"rate_limit","error_details":"429 Too Many Requests"}"#,
)
});
assert!(request.contains(r#""state":"Error""#), "{request}");
assert!(request.contains(r#""detail":"rate_limit""#), "{request}");
}
/// An `error` value that is not a plain enum spelling (or a payload without
/// one at all) must never produce a broken body — no detail is posted.
#[test]
fn script_posts_no_detail_for_an_unexpected_error_value() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["Error"],
&owned_envs(&session_envs(&url)),
br#"{"hook_event_name":"StopFailure","error":"429 \"Too Many\" \\ requests"}"#,
)
});
assert!(request.contains(r#""state":"Error""#), "{request}");
assert!(!request.contains("detail"), "{request}");
}
/// Events whose payload carries no error fields post a bare body — never an
/// empty or stale detail.
#[test]
fn script_posts_no_detail_for_events_without_an_error() {
let request = capture_request(|url| {
sh(
HOOK_SCRIPT,
&["AwaitingInput"],
&owned_envs(&session_envs(&url)),
br#"{"hook_event_name":"Stop","stop_hook_active":false}"#,
)
});
assert!(request.contains(r#""state":"AwaitingInput""#), "{request}");
assert!(!request.contains("detail"), "{request}");
}
/// A dead or unreachable daemon must never fail — or slow down — a session.
@@ -104,6 +167,7 @@ fn script_ignores_daemon_failure() {
("MAESTRO_URL", "http://127.0.0.1:1"),
("MAESTRO_SESSION_ID", "00000000-0000-0000-0000-000000000000"),
],
br#"{"session_id":"x"}"#,
);
assert!(out.status.success());
}
@@ -138,9 +202,10 @@ fn fresh_install_registers_every_event() {
install(&settings_path(home), &script_path(home)).unwrap();
let cmds = installed_commands(&settings_path(home));
assert_eq!(cmds.len(), 3);
assert_eq!(cmds.len(), HOOK_EVENTS.len());
let joined = cmds.join("\n");
assert!(joined.contains("Active"));
assert_eq!(joined.matches("Active").count(), 2);
assert!(joined.contains("Error"));
assert_eq!(joined.matches("AwaitingInput").count(), 2);
for cmd in &cmds {
assert!(cmd.contains(SCRIPT_FILE_NAME));
@@ -158,7 +223,10 @@ fn install_is_idempotent() {
install(&settings_path(home), &script_path(home)).unwrap();
let twice = std::fs::read_to_string(settings_path(home)).unwrap();
assert_eq!(once, twice);
assert_eq!(installed_commands(&settings_path(home)).len(), 3);
assert_eq!(
installed_commands(&settings_path(home)).len(),
HOOK_EVENTS.len()
);
}
/// An older maestro entry under the same events is updated in place, not
@@ -188,7 +256,11 @@ fn install_updates_an_older_maestro_hook_in_place() {
install(&settings, &new_script).unwrap();
let cmds = installed_commands(&settings);
assert_eq!(cmds.len(), 3, "old entry replaced, not kept: {cmds:?}");
assert_eq!(
cmds.len(),
HOOK_EVENTS.len(),
"old entry replaced, not kept: {cmds:?}"
);
assert!(!cmds.iter().any(|c| c == old_script));
let root: serde_json::Value =
serde_json::from_str(&std::fs::read_to_string(&settings).unwrap()).unwrap();
@@ -231,11 +303,11 @@ fn install_preserves_foreign_hooks_and_removes_stale_maestro_ones() {
assert_eq!(pre[0]["hooks"][0]["command"], "my-own-linter.sh");
// Stale maestro event gone.
assert!(root["hooks"].get("UserPromptSubmit").is_none());
// Our three events present, nothing else maestro-related anywhere.
// Our events present, nothing else maestro-related anywhere.
let cmds = installed_commands(&settings);
assert_eq!(
cmds.iter().filter(|c| c.contains(HOOK_MARKER)).count(),
3,
HOOK_EVENTS.len(),
"{cmds:?}"
);
}
+37 -12
View File
@@ -19,8 +19,7 @@ use tracing_subscriber::EnvFilter;
#[command(
name = "maestro",
version,
about = "Unified CLI / daemon / TUI for the Maestro terminal-session orchestrator",
arg_required_else_help = true
about = "Unified CLI / daemon / TUI for the Maestro terminal-session orchestrator"
)]
struct Cli {
/// Override the daemon HTTP base clients talk to (e.g.
@@ -37,8 +36,11 @@ struct Cli {
#[arg(short, long, global = true, action = clap::ArgAction::Count)]
verbose: u8,
/// A missing subcommand defaults to the TUI (#119): a bare `maestro` opens
/// the terminal UI. Builds without the `tui` feature have no UI surface,
/// so they print help and exit 0 instead.
#[command(subcommand)]
command: Command,
command: Option<Command>,
}
#[derive(Debug, Subcommand)]
@@ -89,7 +91,22 @@ fn main() -> ExitCode {
}
async fn dispatch(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
match cli.command {
let command = match cli.command {
Some(command) => command,
None => {
#[cfg(feature = "tui")]
{
Command::Tui(maestro_tui::cli::TuiArgs { daemon_url: None })
}
#[cfg(not(feature = "tui"))]
{
use clap::CommandFactory;
let _ = <Cli as CommandFactory>::command().print_help();
return Ok(());
}
}
};
match command {
#[cfg(feature = "cli")]
Command::List(args) => {
maestro_cli::dispatch(
@@ -166,19 +183,19 @@ fn init_tracing(cli: &Cli) {
}
#[cfg(feature = "cli")]
fn is_cli_subcommand(command: &Command) -> bool {
fn is_cli_subcommand(command: &Option<Command>) -> bool {
matches!(
command,
Command::List(_)
| Command::Spawn(_)
| Command::Attach(_)
| Command::Kill(_)
| Command::Hooks(_)
Some(Command::List(_))
| Some(Command::Spawn(_))
| Some(Command::Attach(_))
| Some(Command::Kill(_))
| Some(Command::Hooks(_))
)
}
#[cfg(not(feature = "cli"))]
fn is_cli_subcommand(_command: &Command) -> bool {
fn is_cli_subcommand(_command: &Option<Command>) -> bool {
false
}
@@ -216,12 +233,20 @@ mod tests {
}
/// Parse argv the way `main` does, returning just the dispatched subcommand.
fn command_of(argv: &[&str]) -> Command {
fn command_of(argv: &[&str]) -> Option<Command> {
Cli::try_parse_from(argv)
.unwrap_or_else(|e| panic!("parse {argv:?}: {e}"))
.command
}
#[test]
fn bare_invocation_parses_to_no_subcommand() {
// The TUI-vs-help default is applied in `dispatch`; parsing must not
// reject an empty argv.
assert!(command_of(&["maestro"]).is_none());
assert!(command_of(&["maestro", "--verbose"]).is_none());
}
#[cfg(feature = "cli")]
#[test]
fn cli_subcommands_are_classified_as_cli() {
+27 -3
View File
@@ -54,11 +54,35 @@ fn version_flag_succeeds() {
assert!(text.contains("maestro"), "--version output: {text}");
}
/// A bare `maestro` must never fall back to help on a build that has a TUI:
/// the missing subcommand defaults to `tui` (#119). Point at a port with
/// nothing listening so the TUI's health probe fails fast instead of blocking
/// on a terminal.
#[cfg(feature = "tui")]
#[test]
fn no_args_prints_help_and_fails() {
// `arg_required_else_help = true` => exit non-zero but show usage.
fn bare_invocation_dispatches_to_the_tui() {
let (ok, text) = run(&["--daemon-url", "http://127.0.0.1:9"]);
assert!(
!ok,
"TUI against a dead daemon should exit non-zero:\n{text}"
);
assert!(
text.contains("daemon unreachable"),
"bare invocation should have run the TUI's daemon probe:\n{text}"
);
assert!(
!text.contains("Usage"),
"bare invocation must not print help when a TUI exists:\n{text}"
);
}
/// Builds without the `tui` feature keep the historical behaviour: no args =>
/// help, exit 0 (#119).
#[cfg(not(feature = "tui"))]
#[test]
fn bare_invocation_prints_help_and_exits_zero() {
let (ok, text) = run(&[]);
assert!(!ok, "bare invocation should fail:\n{text}");
assert!(ok, "bare invocation should exit 0:\n{text}");
assert!(text.contains("Usage"), "expected usage text:\n{text}");
}
+3
View File
@@ -119,6 +119,9 @@ The wizard always installs the default-feature build.
`maestro tui` is the keyboard-and-mouse driven terminal UI client. It ships
in the default feature set, so a vanilla `cargo install maestro` builds it.
A bare `maestro` opens the TUI too — the missing subcommand defaults to `tui`
(#119), so `maestro --daemon-url http://host:7070` works as well. Builds
without the `tui` feature have no UI to open and print help instead (exit 0).
Drop the `tui` feature to skip it on hosts that don't need the UI surface
(`--no-default-features --features cli,daemon`).
+14
View File
@@ -107,6 +107,20 @@ When the TTL expires, inference resumes and the session's reported detail —
if any — clears with the report. Clients never blend the two: whichever source
is in force supplies both the colour and the clock for `activity_since`.
Two rulings specific to a reported `Error` (#114):
- **The five-minute TTL applies to `Error` like to any other state.** A
session that failed is still failed after five minutes, but a card pinned
red forever lies once the session moves on; when the report lapses the card
returns to the daemon's own inference — usually `Idle` for a session that
has gone quiet.
- **Only a new report clears an `Error` before its TTL.** The next
`ActivityState` of any kind replaces it (with #106's hook that is typically
`PreToolUse` → `Active` when the user prompts again). PTY output and
process changes do not clear it — only the agent itself may withdraw what
the agent itself reported. Re-reporting `Error` extends the TTL rather than
stacking reports.
If a client receives an `ActivityState` variant it does not know, it must not
fail rendering: treat it as `Idle` (neutral/dim, line-2 rules apply as usual)
until the daemon sends a variant it does know. Rust clients deserialize