sync batch-datagen memory L13 to pid-file launch contract

Overview / Motivation

Auto-filed by the /daily Step-C parked-candidate routing pass (2026-07-09) from a recursion-guard-parked workflow-fix candidate raised on task #1200.

Goal

Add setsid to the L13 launch recipe and point the pid capture at the launcher-internal echo $$ > / atomic tmp+mv contract forms.

Workflow gap

• Bug observed: L13 ('How to apply' step 1) prescribes a nohup launch recipe lacking setsid and with an unspecified pid-capture method — after #1200 it is the last experimenter memory prescribing a non-trio launch (verified still present on main). #1200's acceptance grep did not catch it (no echo $! > literal); pre-existing and outside #1200's fenced two-file scope.
• Why it is a workflow gap: Agent memories are always-loaded workflow surface steering the experimenter; a memory prescribing a non-setsid launch recipe will be copied verbatim, re-introducing the SIGHUP-reap class the pid-file launch contract (pod-side-reporting.md, #1070) closed.
• Confidence (emitter): medium

Proposed change (candidate diff sketch — refine in planning)

• 1. Launch with full nohup ... < /dev/null > log 2>&1 &, capture the PID, ...
  • 1. Launch with full setsid nohup ... < /dev/null > log 2>&1 &, capture the PID per the pid-file launch contract (launcher-internal echo $$ > pre-exec, or atomic tmp+mv), ...

Scope / surfaces

• Primary target: .claude/agent-memory/experimenter/feedback_datagen_anthropic_batch_long_running.md

Constraints / invariants

• Workflow-surface only. scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
• This session runs under EPM_WORKFLOW_FIX_SESSION=1 semantics (workflow_fix_target Provenance line) — it MUST NOT auto-route its own subagents' workflow-fix candidates.

Provenance

• workflow_fix_target: .claude/agent-memory/experimenter/feedback_datagen_anthropic_batch_long_running.md
• fingerprint: 35e9ec429bb9

Workflow-fix candidate received from Alternatives critic (Phase 2, plan v1 review) — PARKED, NOT ROUTED: this session runs under the workflow-fix recursion guard (task #1200 carries a workflow_fix_target: Provenance line; see .claude/rules/workflow-fix-on-bug.md § Recursion guard). routed: parked: EPM_WORKFLOW_FIX_SESSION. Next human/orchestrator pass may file it. source: prose-followup

target_file: .claude/agent-memory/experimenter/feedback_datagen_anthropic_batch_long_running.md bug_observed: L13 PRESCRIBES a launch recipe ("Launch with full nohup ... < /dev/null > log 2>&1 &, capture the PID") lacking setsid and with an unspecified pid-capture method — after #1200 lands it is the last experimenter memory prescribing a non-trio launch. No echo $! > literal, so #1200's acceptance grep does not catch it; pre-existing and outside #1200's fenced two-file scope. why_workflow_gap: Agent memories are always-loaded workflow surface steering the experimenter; a memory prescribing a non-setsid launch recipe will be copied verbatim, re-introducing the SIGHUP-reap class the pid-file launch contract (pod-side-reporting.md, #1070) closed. proposed_change: Add setsid to the L13 launch recipe and point the pid capture at the launcher-internal echo $$ > / atomic tmp+mv contract forms. diff_sketch: |

• 1. Launch with full nohup ... < /dev/null > log 2>&1 &, capture the PID, ...

• 1. Launch with full setsid nohup ... < /dev/null > log 2>&1 &, capture the PID per the pid-file launch contract (launcher-internal echo $$ > pre-exec, or atomic tmp+mv), ... confidence: medium related_task: #1200