obs(sdk): info-level diagnostics on the pause-cascade hot path

[AbirAbbas]

Summary

Promote five points on the cross-reasoner pause-propagation hot path from debug to info so the next failing run is diagnosable from logs alone — no behavior change.

Why

Run run_1778458437977_fb6f588b (production): implement_from_issue failed with "Reasoner 'implement_from_issue' timed out after 7200.0s of active time" at wall-clock duration_ms=7201393. Its child swe-planner.build was visibly WAITING on a hax-sdk approval for ≥20 min (gap between plan completing at 00:42:34 and the revision iteration of run_architect starting at 01:02:57). Math: if the cascade had fired, pause_clock.total_paused() ≥ 1200s → active_elapsed ≤ 6001s → watchdog wouldn't have tripped. It tripped at 7201s wallclock, so pause_clock.total_paused() ≈ 0 — the cascade never started.

A local 3-hop repro (outer→middle→inner-with-app.pause) at the same SDK version (0.1.83) does fire the cascade correctly (outer's CP exec went to waiting / awaiting_child within seconds, survived 30s past a 5s budget). Something differs in production. All cascade toggles were emitted at logger.debug level, so the production log was silent on which specific link broke — no signal on whether parent_pause_clock lookup missed, whether the polling task ever observed the awaited child as WAITING, or whether pause_clock.start_pause() fired.

What this PR adds

All info level (so they appear in production without dev_mode), all on hot path but bounded — at most one line per state transition.

1. agent.py — _execute_async_with_callback: log pause_clock registration with id() + execution_id + reasoner_name at reasoner entry.
2. agent.py — app.call (cross-reasoner pause prop block): log target, parent_execution_id, parent_pause_clock id(), and a peek at _pause_clocks keys. Distinguishes "no current_context" from "lookup missed" — they need different fixes.
3. agent.py — watchdog firing: log wall_elapsed, total_paused, active_elapsed, budget, pause_clock_id at the moment the timeout cancels the reasoner. Tells "legitimate long active work" apart from "cascade never ran" without re-deriving from the run timeline.
4. async_execution_manager.py — wait_for_result pause toggles: log start_pause and end_pause on the parent's pause_clock with the awaited child id + clock id + cumulative paused seconds.
5. async_execution_manager.py — poll status mapper: log WAITING/RUNNING transitions on the awaited child at info (other transitions stay at debug to avoid spam from healthy workflows).

Diagnostic value

The next production run that hits this failure mode will show, in order:

• (1) pause_cascade: registered pause_clock id=<X> for execution_id=<parent>
• (2) pause_cascade: target=<child> parent_execution_id=<parent> parent_pause_clock_id=<X> (or None — which itself names the bug)
• (5) pause_cascade: poll_observed execution_id=<child> status_transition=running->waiting (or missing — which names a different bug)
• (4) pause_cascade: start_pause child=<child> pause_clock_id=<X> (or missing — names yet another bug)
• (3) pause_cascade: WATCHDOG_FIRING ... total_paused=<N>s active_elapsed=<N>s budget=<N>s

Diff between expected-vs-actual at each step pinpoints the broken link.

Validation Contract

• No behavior change. Same imports, same control flow, same condition checks.
• One additional log_info at reasoner start (per execution).
• One additional log_info per app.call from a reasoner (per call).
• One additional logger.info per WAITING/RUNNING transition on awaited child (bounded by the number of child status flips — typically 2-3 per pause cycle).
• One additional log_error if and only if the watchdog actually fires (rare by design).

Test plan

• ruff check . — clean
• python -m pytest --no-cov — passes (exit 0)
• Deploy to Railway test stack
• Trigger another implement_from_issue run that pauses on hax approval
• Read pause_cascade: log lines on github-buddy to identify which link broke

[github-actions]

Performance

SDK	Memory	Δ	Latency	Δ	Tests	Status
Python	9.4 KB	+4%	0.32 µs	-9%	✓	✓

✓ No regressions detected

[github-actions]

📊 Coverage gate

Thresholds from .coverage-gate.toml: per-surface ≥ 86%, aggregate ≥ 88%, max per-surface regression ≤ 1.0 pp, max aggregate regression ≤ 0.50 pp.

Surface	Current	Baseline	Δ
control-plane	87.50%	87.30%	↑ +0.20 pp	🟡
sdk-go	91.90%	90.70%	↑ +1.20 pp	🟢
sdk-python	93.73%	93.63%	↑ +0.10 pp	🟢
sdk-typescript	92.68%	92.56%	↑ +0.12 pp	🟢
web-ui	89.91%	90.01%	↓ -0.10 pp	🟡
aggregate	89.02%	89.01%	↑ +0.01 pp	🟡

✅ Gate passed

No surface regressed past the allowed threshold and the aggregate stayed above the floor.

[github-actions]

📐 Patch coverage gate

Threshold: 80% on lines this PR touches vs origin/main (from .coverage-gate.toml:thresholds.min_patch).

Surface	Touched lines	Patch coverage	Status
control-plane	0	—	➖ no changes
sdk-go	0	—	➖ no changes
sdk-python	0	—	➖ no changes
sdk-typescript	0	—	➖ no changes
web-ui	0	—	➖ no changes

✅ Patch gate passed

Every surface whose lines were touched by this PR has patch coverage at or above the threshold.