workflow-fix: reap/retry half-spawned session on daemon webhook timeout

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised during chat-mode orchestration (spawning the #952 session; emitting agent: orchestrator).

Goal

On a webhook-timeout HTTP 500 from the daemon /spawn-session, spawn_session.py stops (or retries the handshake for) the already-forked half-spawned session instead of leaving an idle unmapped session running.

Workflow gap

• Bug observed: spawn-issue --issue 952 --auto returned HTTP 500 Session webhook timeout for PID <n> twice (PIDs 3698, 4687) while the Happy daemon had ALREADY forked live claude sessions for both attempts — leaking two live-but-empty unmapped sessions (no initial prompt delivered, no issue mapping, no crash-recovery registration) that required manual spawn_session.py stop. A third identical invocation succeeded, so the timeout is load-transient, but every timed-out attempt leaks a session.
• Why it is a workflow gap: the spawn helper reports failure while leaving daemon-side state half-created; nothing cleans it up (the watcher's idle-unmapped pass only reaps after ≥12h transcript idle), and the caller has no way to know a session leaked. The kickoff path is also missing any retry/backoff for this known-transient daemon error.
• Confidence (emitter): high

Proposed change (candidate diff sketch — refine in planning)

In scripts/spawn_session.py spawn-issue/spawn-pm/spawn-campaign error path:
+ on HTTP 500 whose error matches 'Session webhook timeout for PID (\d+)':
+   1. re-query the daemon session list for a session whose pid == captured PID
+   2. if found: either (a) complete the registration (deliver initial prompt +
+      issue mapping + crash-recovery registration) if the daemon API allows, or
+      (b) stop the half-spawned session before raising, so a failed spawn
+      leaves no orphan
+   3. retry the spawn once with backoff (the error is load-transient; a
+      third manual attempt succeeded on 2026-07-03)
+ log the orphan-reap/retry decision to stderr so callers see it

Scope / surfaces

• Primary target: scripts/spawn_session.py
• Grep the workflow surface for the pattern before editing (grep -rln 'webhook timeout\|spawn-session' .claude/ CLAUDE.md scripts/) and update every hit; list them in the plan. Check whether autonomous_session_watch.py's idle-unmapped pass should also learn a fast-path for zero-transcript sessions younger than 12h whose spawn was never registered.

Constraints / invariants

• Workflow-surface only — never experiment code, configs/, or tasks/.
• scripts/workflow_lint.py --check-asks passes; ruff on touched files passes; if workflow.yaml or CLAUDE.md change, they stay consistent with the rule file.
• This session runs under EPM_WORKFLOW_FIX_SESSION=1 and carries a workflow_fix_target: Provenance line — it MUST NOT auto-route any of its own subagents' workflow-fix candidates (recursion guard).

Provenance

• workflow_fix_target: scripts/spawn_session.py
• fingerprint: 547d7b5f5d26

Raised from chat-mode orchestration 2026-07-03 while spawning the #952 autonomous session: two consecutive spawn-issue --issue 952 --auto calls failed with Happy daemon /spawn-session returned HTTP 500: {'success': False, 'error': 'Session webhook timeout for PID 3698'} (then PID 4687); spawn_session.py list showed both PIDs as live running sessions with no issue mapping; both were stopped manually; the third attempt succeeded (session cmr5tu41o1uagwc0uchgi7huy).