daily-fix: codex_task result-fetch retry + job re-attach

Overview / Motivation

Auto-filed by the /daily 2026-07-02 backfill problem sweep (route 2 — behavior/logic change, independent review required). Emitting context: transcript mining of the 2026-07-02 sessions.

Goal

Retry the result-fetch before declaring a Codex job lost, and support re-attach to an already-spawned job after a wrapper kill instead of re-running the whole job

Workflow gap

• Bug observed: On 2026-07-02 at least 7 Codex dispatches failed transiently across sessions: result-fetch reported 'No job found' while the job had run (exit 5/7/8), and wrapper kills left orphaned jobs (exit 4), each costing a full re-dispatch
• Why it is a workflow gap: codex_task.py drops orphaned markers on marker-post failure but has no result-fetch retry or job re-attach path, so every transient runtime blip re-pays the whole Codex run in latency and compute
• Confidence (emitter): medium

Proposed change (candidate diff sketch — refine in planning)

+ on 'No job found': re-probe the job registry 2-3x with backoff before
+   declaring lost (distinguish job-unknown from result-not-yet-written)
+ add --reattach <job_id>: skip spawn, poll an existing job to completion
+   and write the output file (wrapper-kill recovery)

Scope / surfaces

• Primary target: scripts/codex_task.py
• Grep the workflow surface for the pattern before editing and update every hit; list them in the plan.

Constraints / invariants

• Workflow-surface only — never experiment code, configs/, or tasks/.
• scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
• 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/codex_task.py
• fingerprint: b4a37002a759
• related_task: #830
• source: /daily 2026-07-02 backfill problem sweep (route 2)