workflow-fix: canonicalize setsid launcher-script SSH-MCP pod launch shape

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a prose follow-up surfaced on task #841 (emitting agent: experimenter).

Goal

Canonicalize the setsid launcher-script launch shape (sh-safe dot-source of .env, setsid + </dev/null, launcher writes its own pidfile) as the SSH-MCP pod launch pattern in experimenter.md and the SKILL.md Step 6d.1 brief template.

Workflow gap

• Bug observed: the Step 6d.1 brief's literal source .env && nohup bash -c '...' & echo $! launch shape fails under the SSH-MCP sh shell (source: not found, incident #545) and risks SIGHUP reaping without setsid/</dev/null (#444/#541); the #841 experimenter had to improvise the safe shape at launch time.
• Why it is a workflow gap: the launch-command shape is prescribed by the workflow surface (the SKILL.md Step 6d.1 brief template quotes "the exact nohup launch command"; experimenter.md owns the launch protocol) — every future experimenter either re-hits the sh-shell/source failure or silently deviates from its brief.
• Confidence (emitter): high (the #841 experimenter hit and worked around it live; #545 is the recorded prior incident).

Proposed change (candidate diff sketch — refine in planning)

• experimenter.md launch protocol: replace the bare nohup ... & echo $! prescription with the launcher-script pattern:
  • write /workspace/launch_issue_.sh: #!/bin/sh … . ./.env (dot-source, sh-safe) … setsid nohup bash -c '<chain>' > <log> 2>&1 < /dev/null & echo $! > <pidfile>
  • launch via sh /workspace/launch_issue_<N>.sh; verify pid + log advancing as today.
• SKILL.md Step 6d.1 brief template: reference the launcher-script shape (not a literal top-level source .env && nohup line).

Scope / surfaces

• Primary target: .claude/agents/experimenter.md, .claude/skills/issue/SKILL.md
• Grep the workflow surface for the pattern before editing (grep -rln 'nohup' .claude/ CLAUDE.md scripts/bootstrap_pod.sh) and update every prescriptive launch-shape 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: .claude/agents/experimenter.md, .claude/skills/issue/SKILL.md
• fingerprint: c202f67dbb08

Surfaced prose (verbatim, from the #841 experimenter's launch report): "One deviation from the brief's launch shape, same logical contract: I wrapped the three-script chain in a setsid-launcher script (/workspace/launch_issue_841.sh) that sources .env internally via . ./.env and writes its own pidfile, rather than the brief's literal top-level source .env && nohup bash -c '...' & line. Reason: the SSH-MCP shell is sh, so a top-level source .env fails (sh: source: not found, incident #545) and the captured $! would catch a stray job; and bare nohup ... & without setsid/</dev/null risks SIGHUP reaping on session exit (#444/#541). … For future briefs, the setsid-launcher-script pattern is the safe SSH-MCP launch shape."