workflow-fix: index.lock-resilient _run_git + crash-safe set_status registry move

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised on task #825 (emitting agent: orchestrator).

Goal

Make task_workflow git operations resilient to concurrent .git/index.lock collisions and make set_status's folder-move + registry update crash-safe.

Workflow gap

• Bug observed: task.py set-status 825 verifying interpreting crashed with CalledProcessError (git add exit 128) while a concurrent session held .git/index.lock, leaving a PARTIAL mutation: the task folder was already moved on disk to tasks/interpreting/825 but REGISTRY.json still pointed at tasks/verifying/825 and no commit landed. Every subsequent task.py find/view 825 raised FileNotFoundError until a manual task.py audit --repair --apply.
• Why it is a workflow gap: CLAUDE.md § "Concurrent repo-root committers" instructs HUMANS to "retry once on an index.lock collision", but task_workflow._run_git has no such retry, and set_status performs the disk move and the registry/commit non-atomically — a mid-mutation crash strands the task in an unresolvable state on a fleet where many sessions commit concurrently (this WILL recur).
• Confidence (emitter): high

Proposed change (candidate diff sketch — refine in planning)

+ # task_workflow._run_git: on CalledProcessError with stderr matching
+ # 'index.lock|Unable to create .* File exists', sleep 2-5s and retry ONCE.
+ # set_status: perform the registry update in the same critical section as
+ #   the move; on any git failure after the move, roll the move back OR
+ #   complete the registry update before raising (no stranded partial state).
+ # find_task_path: when the registry path is missing on disk, fall back to a
+ #   one-shot on-disk scan (tasks/*/<N>) before raising, and log the drift.

Scope / surfaces

• Primary target: src/explore_persona_space/task_workflow.py
• Grep the workflow surface for the pattern before editing (grep -rn '_run_git\|index.lock' src/explore_persona_space/task_workflow.py scripts/) 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: src/explore_persona_space/task_workflow.py
• fingerprint: 5fc285801c9c

target_file: src/explore_persona_space/task_workflow.py bug_observed: task.py set-status crashed with git add exit 128 on a concurrent .git/index.lock, leaving folder moved but REGISTRY.json stale; find/view raised FileNotFoundError until manual audit --repair --apply why_workflow_gap: _run_git has no retry on the well-known index.lock collision class (CLAUDE.md instructs humans to retry once) and set_status's move+registry+commit is not crash-safe on a fleet of concurrent committers proposed_change: retry git commands once on index.lock collision in _run_git and make set_status registry-move crash-safe (registry-stale fallback in find_task_path) diff_sketch: |

• _run_git: retry once (2-5s backoff) on exit-128 index.lock errors
• set_status: registry update + move in one critical section; no stranded partial state
• find_task_path: on-disk tasks/*/ fallback when registry path missing confidence: high related_task: #825