orchestrate/SKILL.md — Mental queue for /pr-test serialization is fragile across context resets

[vibemasshq-dev]

Hi team,
While auditing this repo with VibeMass I noticed the orchestrate SKILL.md relies on Claude's in-context "mental queue" to serialize /pr-test runs and prevent port conflicts and database corruption.
This pattern works within a single session but breaks silently when:

• Context window compacts mid-orchestration

• Session restarts between PR tests

• Multiple orchestrator instances run simultaneously

A lightweight mechanical guard would make this more robust without changing the workflow:

# At start of /pr-test
if [ -f /tmp/pr-test.lock ]; then
  echo "Another pr-test is running. Please wait."
  exit 1
fi
touch /tmp/pr-test.lock
trap "rm -f /tmp/pr-test.lock" EXIT

This could be added to the SKILL.md as a recommended pattern alongside the existing mental queue instruction.
Happy to open a PR if this direction makes sense to the team.