feat(control) 5/5: a person in a session no longer fails a dispatch - #185
Conversation
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
sanil-23 has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.
📝 WalkthroughWalkthroughThe daemon now emits unthrottled hold and resume markers and allows higher default concurrency. The hub pauses watchdogs during holds. The TUI queues conflicting work, suspends active turns during takeover, resumes them after hand-back, and blocks takeover of remote sessions. ChangesSession control flow
Estimated code review effort: 5 (Critical) | ~120 minutes Sequence Diagram(s)sequenceDiagram
participant Dispatcher
participant PtyManager
participant PtySessionExecutor
participant Operator
Dispatcher->>PtyManager: inspect workspace sessions
PtyManager-->>Dispatcher: return session candidates
Dispatcher->>PtySessionExecutor: queue or start task
Operator->>PtySessionExecutor: take over active session
PtySessionExecutor->>Operator: emit held marker
Operator->>PtySessionExecutor: return control
PtySessionExecutor->>PtySessionExecutor: inject hand-back prompt
Possibly related PRs
Suggested reviewers: Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches📝 Generate docstrings
Comment |
1151aaf to
bbd619c
Compare
There was a problem hiding this comment.
sanil-23 has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.
|
@coderabbitai review |
|
|
@coderabbitai review |
|
bbd619c to
c133b60
Compare
There was a problem hiding this comment.
sanil-23 has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.
|
@coderabbitai review |
|
Phase E of the agent topology: dispatch candidacy and the control
semantics around a session an operator is holding. Everything here is
exercisable on one machine — declare two agents, dispatch, take a
session, watch the work route around it, hand back, see the result.
Candidacy (E1) is now two independent rules that only coincide today.
A hold is on a *session*: `claim_idle` already refuses to hand a
user-owned session to the orchestrator, so reuse is consulted first and
a person working never makes a dispatch fail. Serialization is on the
*checkout*: under `strategy: checkout` an agent's sessions share one
working tree, so a fresh session cannot start beside the writer that is
there — it queues. `PtyManager::operator_hold(cwd)` ("is this workspace
held") is replaced by `sessions_in(cwd)`, a neutral query, with the
policy stated where the strategy lives. Under `worktree` (§G) rule 2
simply stops applying, with no control logic to revisit. F3 still owns
serializing two *orchestrator* sessions in one checkout, which main
allows and which is a scheduling change, not a control one.
A mid-turn takeover suspends instead of discarding (E3): the fold, its
events, usage and workspace context are retained, everything already
written is folded out first, and the task stays open. Held time does not
accrue against the worker's idle ceiling, and the worker announces the
hold so the hub's no-progress window pauses too (E4) — gated exactly
like link liveness, so a worker that dies mid-hold is still reaped once
the session comes back. On hand-back the runtime runs a *fresh* turn in
that same session (E5), prompted with the original instruction plus a
directive to review the session history and workspace state and either
report the finished work or complete it; its answer is emitted as the
pending task's result, under the same task id, because it is the same
call. The blanket `harnessHeld` refusal is retired (E6): the only path
left to it is a queue that outlived the caller's budget, which keeps its
frame byte-identical so the backend's retry stays a retry. Taking
control is gated to local sessions and says so by name (E7).
Control state is no longer advertised at all: `control`, `controlReason`,
`controlSince` and the handback brief are per-agent keys describing a
per-session fact, and a backend folding them by `agentId` would mark
every task on an agent as held when a person took one session. The local
state and every behaviour it drives are unchanged; only the wire is
quieter. That makes the hand-back turn the *only* way a held in-flight
task reaches a result, so a dispatch that meets a person now always ends
in a real result or a real error — never in silence.
Co-Authored-By: Claude <noreply@anthropic.com>
…a slot Both the networked and embedded daemons capped concurrent tasks at 2, so a third dispatch waited on a semaphore permit even when it targeted a different agent in a different workspace — where nothing could collide. That cap predates declared agents: it was the only guard when a machine was one worker with one implicit session. The limits that own the real hazard now sit at its grain: per-agent max_sessions derived from the workspace strategy, and the checkout serialization that keeps a second writer out of a tree someone is in. A host-wide count knows about neither, so it could only delay work that was already safe. The semaphore stays as the accounting behind active_count, and an operator can still set concurrency to impose a real cap on a small machine. Co-Authored-By: Claude <noreply@anthropic.com>
c133b60 to
db3ec0c
Compare
There was a problem hiding this comment.
sanil-23 has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.
|
@coderabbitai review |
✅ Action performedReview finished.
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (3)
src/tui/src/ui/app/types.rs (1)
1126-1135: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueKeep the file below 500 lines of code.
src/tui/src/ui/app/types.rscontains 490 non-blank, non-comment lines, so it does not currently violate the 500-line limit. Split the file before future changes exceed the limit.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/tui/src/ui/app/types.rs` around lines 1126 - 1135, Keep src/tui/src/ui/app/types.rs under 500 non-blank, non-comment lines by moving related type definitions or fields, including pane_remote_session, into an appropriate module before adding further content. Update module declarations and references so behavior and visibility remain unchanged.Source: Coding guidelines
src/tui/src/worker/executor_tests/control.rs (1)
283-327: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winAdd coverage for the two abort branches on the hold path.
hold.rsadds two error branches that no test in this module reaches:
await_checkout_releasereturns"task aborted while queued behind an operator"when the requester aborts while the dispatch is parked in the queue.await_handbackreturns"{provider} task aborted while an operator held the session"when the requester aborts during a hold.Both are ordinary events: an orchestrator that gives up on a task while a person is at the keyboard. Neither is pinned, so a regression that ignores the abort and waits out the full budget would pass this suite. The second branch also carries a specific invariant worth asserting — the operator's session must survive the abort, because
finish_turnskipsclosewhilecontrol == User.Each test mirrors an existing one and only needs
opts.abortretained and aborted mid-wait.Do you want me to generate both tests?
As per coding guidelines: "Cover new branches and maintain the 80% line-coverage gate defined by.github/workflows/ci.yml."🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/tui/src/worker/executor_tests/control.rs` around lines 283 - 327, Add two focused tests in control.rs to cover the new abort branches in hold.rs: one for await_checkout_release returning the queued-abort error and one for await_handback returning the held-abort error. Reuse the existing harness, options, and fake_harness_script setup, keep opts.abort enabled, trigger the abort mid-wait, and assert the exact abort message path is reached instead of waiting for the full timeout. In the await_handback case, also assert the operator session still exists after the abort to preserve the finish_turn/control == User behavior.Source: Coding guidelines
src/sdk/src/hub/runner/mod.rs (1)
101-120: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winMove the new behavior out of
mod.rs.Both changes add behavior to directory module roots. Keep these files as module wiring.
src/sdk/src/hub/runner/mod.rs#L101-L120: movelive_sleepinto a focused watchdog submodule and import it frommod.rs.src/sdk/src/hub/runner/pump/mod.rs#L157-L190: move control-marker parsing and status-control handling into a focused pump submodule.As per coding guidelines,
**/mod.rs: “Keepmod.rsfocused on module documentation,mod/pub usewiring, and glue that fits no more specific submodule.”🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/sdk/src/hub/runner/mod.rs` around lines 101 - 120, Move live_sleep from src/sdk/src/hub/runner/mod.rs lines 101-120 into a focused watchdog submodule, then import or re-export it from the runner module root without changing its behavior. Also move the control-marker parsing and status-control handling from src/sdk/src/hub/runner/pump/mod.rs lines 157-190 into a focused pump submodule, wiring it through pump/mod.rs; both module roots should retain only documentation, module declarations, re-exports, and necessary glue.Source: Coding guidelines
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/tui/src/ui/app/render/agents/mod.rs`:
- Around line 129-140: Move the session-selection and remote-session
classification logic from Agents::render in
src/tui/src/ui/app/render/agents/mod.rs:129-140 into a dedicated Agents child
module, preserving its behavior and wiring the module from mod.rs. Also move the
frame-state reset behavior from src/tui/src/ui/app/render/mod.rs:297-299 into a
dedicated render child module; both mod.rs files should retain only
documentation, module/public-use wiring, and unavoidable glue.
In `@src/tui/src/ui/app/session_control_tests.rs`:
- Around line 42-46: Extend the session-control tests around
app.take_session_control to exercise the render-driven lifecycle: create a
remote RailRow::Session, draw the Agents view so remote-session classification
runs, then invoke take_session_control; subsequently configure local sessions,
draw a non-Agents tab to trigger the per-frame reset, and verify the resulting
empty-row behavior is unaffected by stale remote state. Replace or supplement
direct pane_remote_session assignment with the render-based setup.
In `@src/tui/src/worker/executor/hold.rs`:
- Around line 63-74: Update handback_prompt to normalize instruction by
collapsing or replacing newline characters with spaces before interpolating it,
while preserving the existing prompt wording and behavior. Ensure the returned
string remains a single physical line even when the supplied instruction is
multi-line.
---
Nitpick comments:
In `@src/sdk/src/hub/runner/mod.rs`:
- Around line 101-120: Move live_sleep from src/sdk/src/hub/runner/mod.rs lines
101-120 into a focused watchdog submodule, then import or re-export it from the
runner module root without changing its behavior. Also move the control-marker
parsing and status-control handling from src/sdk/src/hub/runner/pump/mod.rs
lines 157-190 into a focused pump submodule, wiring it through pump/mod.rs; both
module roots should retain only documentation, module declarations, re-exports,
and necessary glue.
In `@src/tui/src/ui/app/types.rs`:
- Around line 1126-1135: Keep src/tui/src/ui/app/types.rs under 500 non-blank,
non-comment lines by moving related type definitions or fields, including
pane_remote_session, into an appropriate module before adding further content.
Update module declarations and references so behavior and visibility remain
unchanged.
In `@src/tui/src/worker/executor_tests/control.rs`:
- Around line 283-327: Add two focused tests in control.rs to cover the new
abort branches in hold.rs: one for await_checkout_release returning the
queued-abort error and one for await_handback returning the held-abort error.
Reuse the existing harness, options, and fake_harness_script setup, keep
opts.abort enabled, trigger the abort mid-wait, and assert the exact abort
message path is reached instead of waiting for the full timeout. In the
await_handback case, also assert the operator session still exists after the
abort to preserve the finish_turn/control == User behavior.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 20060893-fc79-4db9-8e36-ff22eac25efd
📒 Files selected for processing (33)
src/sdk/src/daemon/embedded/types.rssrc/sdk/src/daemon/entry.rssrc/sdk/src/daemon/mod.rssrc/sdk/src/daemon/task_loop/run.rssrc/sdk/src/daemon/tests/capability_tests.rssrc/sdk/src/daemon/tests/mod.rssrc/sdk/src/daemon/types.rssrc/sdk/src/hub/roster/mod.rssrc/sdk/src/hub/runner/mod.rssrc/sdk/src/hub/runner/pump/mod.rssrc/sdk/src/hub/runner/types.rssrc/sdk/src/hub/socket/task_run.rssrc/sdk/src/hub/tests/handoff_advert.rssrc/sdk/src/hub/tests/held.rssrc/sdk/src/hub/tests/held_watchdog.rssrc/sdk/src/hub/tests/mod.rssrc/sdk/src/hub/tests/roster.rssrc/tui/src/ui/app/render/agents/mod.rssrc/tui/src/ui/app/render/mod.rssrc/tui/src/ui/app/session_control.rssrc/tui/src/ui/app/session_control_tests.rssrc/tui/src/ui/app/state.rssrc/tui/src/ui/app/types.rssrc/tui/src/worker/executor/hold.rssrc/tui/src/worker/executor/mod.rssrc/tui/src/worker/executor/run.rssrc/tui/src/worker/executor/types.rssrc/tui/src/worker/executor_tests/control.rssrc/tui/src/worker/executor_tests/mod.rssrc/tui/src/worker/executor_tests/sessions.rssrc/tui/src/worker/pty/manager/session.rssrc/tui/src/worker/pty/tests/control.rssrc/tui/tests/e2e_session_takeover.rs
| // A session row this device is not running: watchable, but not takeable | ||
| // (§E7). Recorded here because this is the only place that can tell the | ||
| // difference — one row down the cursor, both cases are a `None` session. | ||
| self.pane_remote_session = match (&selection.session, selection.rows.get(selection.active)) | ||
| { | ||
| (None, Some(RailRow::Session(row))) => Some( | ||
| row.agent_id | ||
| .clone() | ||
| .unwrap_or_else(|| "another host".to_string()), | ||
| ), | ||
| _ => None, | ||
| }; |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | 🏗️ Heavy lift
Move render behavior out of these mod.rs files.
Both changes add implementation logic to mod.rs. Keep each mod.rs limited to module documentation and wiring. Move the behavior into dedicated child modules.
src/tui/src/ui/app/render/agents/mod.rs#L129-L140: move session-selection and remote-session classification into an Agents child module.src/tui/src/ui/app/render/mod.rs#L297-L299: move frame-state reset behavior into a dedicated render child module.
As per coding guidelines, “Keep mod.rs focused on module documentation, mod/pub use wiring, and glue that fits no more specific submodule.”
📍 Affects 2 files
src/tui/src/ui/app/render/agents/mod.rs#L129-L140(this comment)src/tui/src/ui/app/render/mod.rs#L297-L299
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/tui/src/ui/app/render/agents/mod.rs` around lines 129 - 140, Move the
session-selection and remote-session classification logic from Agents::render in
src/tui/src/ui/app/render/agents/mod.rs:129-140 into a dedicated Agents child
module, preserving its behavior and wiring the module from mod.rs. Also move the
frame-state reset behavior from src/tui/src/ui/app/render/mod.rs:297-299 into a
dedicated render child module; both mod.rs files should retain only
documentation, module/public-use wiring, and unavoidable glue.
Source: Coding guidelines
| let mut app = app(); | ||
| app.pane_session = None; | ||
| app.pane_remote_session = Some("mac-studio-claude".to_string()); | ||
|
|
||
| app.take_session_control(); |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Cover the render-driven remote-session lifecycle.
Lines 43-44 set pane_remote_session directly. The tests do not execute remote-row classification in src/tui/src/ui/app/render/agents/mod.rs Lines 132-140 or the per-frame reset in src/tui/src/ui/app/render/mod.rs Lines 297-299.
Add a test that selects a remote RailRow::Session, draws Agents, and invokes take_session_control. Then draw a non-Agents tab with local sessions configured and verify that stale remote state does not affect the empty-row result.
As per coding guidelines, “Cover new branches.”
Also applies to: 82-87
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/tui/src/ui/app/session_control_tests.rs` around lines 42 - 46, Extend the
session-control tests around app.take_session_control to exercise the
render-driven lifecycle: create a remote RailRow::Session, draw the Agents view
so remote-session classification runs, then invoke take_session_control;
subsequently configure local sessions, draw a non-Agents tab to trigger the
per-frame reset, and verify the resulting empty-row behavior is unaffected by
stale remote state. Replace or supplement direct pane_remote_session assignment
with the render-based setup.
Source: Coding guidelines
| /// One line, because it is typed into a composer. A line-oriented harness reads | ||
| /// a prompt up to the first newline, so a multi-line brief would arrive as a | ||
| /// prompt plus stray input. | ||
| pub(super) fn handback_prompt(instruction: &str) -> String { | ||
| format!( | ||
| "An operator took control of this session and has been working in it; \ | ||
| you now have it back. Review this session's history and the current \ | ||
| state of the workspace, then finish this task: {instruction} — if the \ | ||
| work is already done, do not redo it: report the final result. \ | ||
| Otherwise continue from where things now stand and complete it.", | ||
| ) | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Collapse newlines in instruction before interpolating it.
The doc comment states the prompt must be one line, because a line-oriented harness reads a prompt up to the first newline. instruction is options.prompt, which a peer supplies and which is often multi-line. A multi-line instruction therefore splits the hand-back prompt: the harness reads the first physical line as the prompt, and the rest arrives as stray input on the next read. The wording match *"you now have it back"* sits in the first line, so the turn still starts, but it starts against a truncated task description.
Normalize the interpolated value so the invariant the doc states is enforced by the code.
🐛 Proposed fix to keep the prompt on one line
pub(super) fn handback_prompt(instruction: &str) -> String {
+ // The invariant above is enforced here rather than assumed: a delegated
+ // instruction is routinely multi-line, and a raw interpolation would end
+ // the prompt at its first newline and type the remainder as stray input.
+ let instruction = instruction.split_whitespace().collect::<Vec<_>>().join(" ");
format!(
"An operator took control of this session and has been working in it; \
you now have it back. Review this session's history and the current \
state of the workspace, then finish this task: {instruction} — if the \
work is already done, do not redo it: report the final result. \
Otherwise continue from where things now stand and complete it.",
)
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| /// One line, because it is typed into a composer. A line-oriented harness reads | |
| /// a prompt up to the first newline, so a multi-line brief would arrive as a | |
| /// prompt plus stray input. | |
| pub(super) fn handback_prompt(instruction: &str) -> String { | |
| format!( | |
| "An operator took control of this session and has been working in it; \ | |
| you now have it back. Review this session's history and the current \ | |
| state of the workspace, then finish this task: {instruction} — if the \ | |
| work is already done, do not redo it: report the final result. \ | |
| Otherwise continue from where things now stand and complete it.", | |
| ) | |
| } | |
| /// One line, because it is typed into a composer. A line-oriented harness reads | |
| /// a prompt up to the first newline, so a multi-line brief would arrive as a | |
| /// prompt plus stray input. | |
| pub(super) fn handback_prompt(instruction: &str) -> String { | |
| // The invariant above is enforced here rather than assumed: a delegated | |
| // instruction is routinely multi-line, and a raw interpolation would end | |
| // the prompt at its first newline and type the remainder as stray input. | |
| let instruction = instruction.split_whitespace().collect::<Vec<_>>().join(" "); | |
| format!( | |
| "An operator took control of this session and has been working in it; \ | |
| you now have it back. Review this session's history and the current \ | |
| state of the workspace, then finish this task: {instruction} — if the \ | |
| work is already done, do not redo it: report the final result. \ | |
| Otherwise continue from where things now stand and complete it.", | |
| ) | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/tui/src/worker/executor/hold.rs` around lines 63 - 74, Update
handback_prompt to normalize instruction by collapsing or replacing newline
characters with spaces before interpolating it, while preserving the existing
prompt wording and behavior. Ensure the returned string remains a single
physical line even when the supplied instruction is multi-line.
2 participants
Stack 5/5 — base
split/4-wire. Depends on: 1/5, 2/5, 3/5, 4/5.split/1-agent-modelmainsplit/2-tree-uisplit/1-agent-modelHost → Agent → Sessiontreesplit/3-vocabularysplit/2-tree-uisplit/4-wiresplit/3-vocabularyhosts[], per-agenthostId/maxSessions, resultsessionIdsplit/5-controlsplit/4-wireEach PR's diff shows only its own layer.
split/5-control's tree is byte-identical tobcc61e63, the merge commit on #180.Last of five stacked PRs replacing #180. Review 1/5 – 4/5 first. The tip of this branch
is byte-identical to
bcc61e63, the merge commit on #180.What this layer does
feat(control): a person in a session no longer fails a dispatchPhase E of the agent topology: dispatch candidacy and the control semantics around a
session an operator is holding. All of it is exercisable on one machine — declare two
agents, dispatch, take a session, watch the work route around it, hand back, see the
result.
session:
claim_idlealready refuses to hand a user-owned session to theorchestrator, so reuse is consulted first and a person working never makes a dispatch
fail. Serialization is on the checkout: under
strategy: checkoutan agent's sessionsshare one working tree, so a fresh session queues rather than starting beside the writer
already there.
PtyManager::operator_hold(cwd)is replaced by the neutralsessions_in(cwd), with the policy stated where the strategy lives.and workspace context are retained, everything already written is folded out first, and
the task stays open. Held time does not accrue against the worker's idle ceiling, and
the worker announces the hold so the hub's no-progress window pauses too (E4) — gated
exactly like link liveness, so a worker that dies mid-hold is still reaped.
the original instruction plus a directive to review the session history and workspace
state and either report the finished work or complete it. Its answer is emitted as the
pending task's result, under the same task id, because it is the same call.
harnessHeldrefusal is retired (E6): the only path left to it is aqueue that outlived the caller's budget, and that frame stays byte-identical so the
backend's retry stays a retry.
control,controlReason,controlSinceand the handback brief are per-agent keys describing a per-session fact,and a backend folding them by
agentIdwould mark every task on an agent as held when aperson took one session.
perf(daemon): lift the host-wide task cap so work stops queueing for a slotBoth daemons capped concurrent tasks at 2, so a third dispatch waited on a semaphore
permit even when it targeted a different agent in a different workspace. That cap predates
declared agents. The limits that own the real hazard now sit at its grain: per-agent
max_sessionsand checkout serialization. The semaphore stays as the accounting behindactive_count, and an operator can still setconcurrencyto impose a real cap.Reordering note
feat(control)was the eighth commit in the original branch, ahead offix(hub): stop advertising hostId. It moved down one slot so the wire layer (4/5) isself-contained. It replayed with no conflicts and the tip is unchanged.
What a reviewer should look at closely
only if the per-agent and per-checkout limits genuinely cover the hazard the cap used to
cover. Worth checking that an agent with
strategy: checkoutreally cannot get twoconcurrent writers.
held; the gate is deliberately the same as link liveness so a worker that dies mid-hold
is still reaped. That gate is the thing that stops "held" becoming "leaked forever".
dispatch that meets a person always ends in a real result or a real error — never in
silence. Any path that can still drop it is a hang.
them. The local state and every behaviour it drives are unchanged; only the advert is
quieter.
Validation
Result on this branch:
fmt,clippy -D warnings,check, andbuildall clean.testreports 3900 passed, 5 failed, 14 ignored on macOS. All five failures arepre-existing on
mainand unrelated to this stack:daemon::providers::acp::tests::execution::a_new_acp_session_is_reported_before_the_task_completesdaemon::providers::tests::direct_runs_report_the_session_before_workspace_context(both are idle-watchdog timing tests; they pass in isolation)
worker::pty::tests::session::a_launch_root_preserves_trailing_whitespaceworker::pty::tests::session::a_session_snapshots_head_before_the_harness_can_commitworker::pty::tests::session::an_unborn_repository_records_its_root_without_a_launch_commit(the last three are the macOS
/varvs/private/varsymlink)Coverage was deliberately not run locally; CI owns that gate.
Summary by CodeRabbit
New Features
Bug Fixes