runaway: warm-wait check-ins are not empty wakes (GST-1597) (#4069)

* runaway: warm-wait check-ins are not empty wakes (GST-1597)

lib/runaway.ts counted each 240-second warm-wait check-in (docs/captain.md,
#4060) as an empty wake, so /runaway-stopper could flag a lane that follows
the rule. A turn whose only call re-issues the wait on the same operation, or
a 240-second timer, within 30 minutes of the wait that started the run is
now left out of the rate.

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix: pre-landing review fixes

Review-Finding: p1/specialist-batched#1 reader=red-team
Review-Finding: p1/specialist-batched#2 reader=testing
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs: note the warm-wait exclusion in the /runaway-stopper summary; rewrap captain.md (GST-1597 docs audit)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* runaway: keep wait input in the skill collector; 240 s timers chain only on the run's operations; status reads keep the warm window (GST-1597)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

docs/captain.md

@@ -152,10 +152,13 @@ cheap:
   operation with `duration_seconds: 240`, and on a timer wake that brings nothing new it
   re-issues the same wait silently, with no message. Once a warm wait has run about 30
   minutes, or past the end the lane expected, the lane stops re-issuing the timer and
-  sleeps on the operation or pull request watch alone. A wait expected to run longer sleeps
-  on its operations alone. For short lane waits this overrides "waits go on operations,
-  never on polling wakes" ([unattended.md](unattended.md#report-length)). The controller
-  never warm-waits, and a lane still sleeps while /ship runs
+  sleeps on the operation or pull request watch alone. `gstack-runaway scan` leaves these
+  check-ins out of its empty-wake rate: a turn whose only call re-waits on the same
+  operation, or a 240-second timer on it or on nothing, within 30 minutes of the first wait.
+  A status read in between counts and does not restart the 30 minutes. A wait expected to run
+  longer sleeps on its operations alone. For short lane waits this overrides "waits go on
+  operations, never on polling wakes" ([unattended.md](unattended.md#report-length)). The
+  controller never warm-waits, and a lane still sleeps while /ship runs
   ([the lane's /ship sequence](#the-lanes-ship-sequence-before-ready)).
 - **Batched reads.** Independent reads (files, greps, pull request and thread state) go in
   one parallel program, not one call each. Each call returns 4k characters or less; long

docs/skills.md

@@ -1020,7 +1020,7 @@ sets the defaults that keep it from coming back.
 - **Detect.** One program fetches each crew thread (`tools.thread.get`, with its transcript) to files, then
   `gstack-runaway scan --dir <dir>` prints a `RUNAWAY:` line per thread and a `FLAG:` line per limit crossed:
   own tokens past 500M or active past 3 days (rotate), a subagent past 100M (stop), empty wakes past 25% of the
-  last 40 turns, and review rounds past 2 per PR. Captains are reported, never rotated.
+  last 40 turns (warm-wait check-ins are left out), and review rounds past 2 per PR. Captains are reported, never rotated.
 - **Stop.** A live subagent past 100M is stopped with its reason. A thread is rotated at a clean point through
   a handoff file and a successor brief (`runaway-stopper/templates/`), then archived by its own Captain. A thread
   mid-/ship or mid-merge gets `DEFER:` and nothing happens until that work finishes.