Repository navigation
Finite peek --wait marker returns only after the session exits #206
Description
Johannes's assistant (Compounding Manager) reports this from fleet feedback FB-4996.
- Observed:
pty peek --wait "END 004"returned only after the retained session exited, when the marker had already scrolled out of view. A 600 s bounded input window expired before the wait returned. - Expected: a marker printed while the wait is active satisfies the wait at once, without process exit. A missed marker is reported differently from a marker that never appeared. The existing time bound stays.
- Not the same as: session_exit event waits (pty events --wait session_exit busy-loops after target exited and ignores --timeout #196/fix: complete event waits from session lifecycle state #197) or the replay-test scan change in Scan each replay-test packet once while waiting for the marker #176.
Posted on behalf of @schickling
| field | value |
|---|---|
agent_identity |
agent/compoundingtech/compounding-manager |
session |
01a1d992 |
agent_persona |
generalist |
agent_supervisor |
unavailable |
agent_tool |
OMP |
agent_tool_version |
18.12.5 |
agent_runtime |
OMP 18.12.5 |
agent_model |
anthropic/claude-opus-5-5 |
worktree |
unknown |
Activity
myobie commentedon Nov 3, 2026
I reproduced this at main.
peek --waitdoesn't watch output. Every 200 ms it takes a snapshot of the visible screen and looks for the marker there (crates/pty-client/src/peek.rs,peek_wait). If the marker is printed and then scrolled off the screen between two snapshots, no snapshot ever contains it. The wait then runs until the timeout, or, with no time limit, until the session exits and its savedlastLineshappen to include the marker. That is the report's "returned only after the session exited".The repro: a session that waits 1 s, prints
END 004, and then prints 200 more lines at once. The wait ispty peek --plain --wait 'END 004' -t 6.pty peek --fullshows the marker in scrollback.The fix I'm making:
peek --waitkeeps one read-only peek connection open for the whole wait. It matches against the screen snapshot that connection starts with, and then against each output chunk as it arrives, with ANSI stripped. A tail is kept across chunks, so a marker split between two chunks still matches. A marker printed while the wait is active therefore matches as soon as its bytes arrive, whether or not it is still on screen.-tkeeps its meaning.None of this touches tier-1 files. I'll link the PR here.
myobie commentedon Nov 3, 2026
The fix is in #208. With it, the repro above returns at 1.0 s with exit 0 for the scrolled marker, the same as for a marker that stays on screen. A marker from before the wait now times out with a second line: "It is in the scrollback from before the wait began, no longer on screen." A pattern that never appeared keeps the single line it had.
Merge pull request #208 from compoundingtech/agent/peek-wait-watches-…
myobie commentedon Nov 3, 2026
Fixed by #208, merged as d618d49.
peek --waitnow also watches the output while it waits. A marker that is printed and scrolled away during the wait ends the wait at once. On a timeout, a marker that is in the scrollback from before the wait gets a second line, so it reads differently from one that never appeared. The-tbound holds even under output that never pauses.