[chanolan20]

Summary

The full pytest suite is green on Linux CI but failed 82 tests on a stock macOS host (81 failed, 14116 passed before; 14185 passed, 96 skipped, 0 failed after this change). Tracked as #7006. All failures were harness-side; four root causes:

1. basetemp outside the confinement allowlist — tests/helpers/worker_runtime.py set pytest's basetemp to <runtime>/pytest while the private TMPDIR was <runtime>/tmp. On Linux this passes only because the runtime root itself lives under /tmp (an allowlisted root); on macOS the runtime root is /var/folders/..., so every tmp_path path was rejected by _resolve_tool_path (poisoning tool-path-confinement, write_file race, and edit_file tests). The basetemp now lives inside the private TMPDIR.
2. AF_UNIX overflow — macOS's 49-char /var/folders TMPDIR left too few of the 104 sun_path bytes for the tmux socket and agent runtime socket bound under tmp_path. The runtime root now sits at the short real temp root (/private/tmp, 0700 subdir), matching the doctrine already documented in tests/helpers/unix_sockets.py.
3. Enforcing-containment tests on hosts that refuse uncontained runs — 16 tests drive the bwrap filesystem+process_tree enforcing path, which by design refuses to run on hosts without bubblewrap (related to Native macOS (Apple Silicon): bash/python tools refuse to run — "containment unavailable (filesystem, process_tree)" #6688). They are gated by tests/helpers/containment_gate.py, a cached probe of the same mechanism table src/containment.py uses — so they still execute on Linux CI and skip (not fail) elsewhere.
4. Resolved-vs-unresolved temp path comparisons — the exact family documented in tests/KNOWN_FAILURES.md (/tmp → /private/tmp): fixed in test_workspace_confine, test_write_file_empty_body (fixture now yields the realpath), test_review_regressions. The pinned-transport SSRF fallback test gained a probe gate for hosts (stock macOS) where 127.0.0.2 drops instead of refusing.

Target branch

• This PR targets the correct integration branch: dev in the public repository.

Linked Issue

Fixes #7006 (related: #6688)

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)
• New feature (non-breaking — adds new behaviour)
• Breaking change (changes or removes existing behaviour)
• Refactor / cleanup (behaviour unchanged)
• Documentation only
• CI / tooling / configuration

Checklist

• I searched open issues and open PRs — this is not a duplicate.
• This PR targets the correct integration branch (lab in maintainer-preview; dev in the public repository)
• My changes are limited to the scope described above — no unrelated refactors or whitespace changes mixed in.
• I actually ran the app (docker compose up or uvicorn app:app) and verified the change works end-to-end. Type-checks and unit tests are not enough.
• I did not run the app/runtime validation and stated that gap in How to Test. Leave this unchecked when the app-run box above is checked.

How to Test

The change is test-harness-only; there is no app behavior to view in a browser, so the app-run box is intentionally unchecked. To verify:

1. On any Linux machine (or in CI — the gated tests run unchanged where bubblewrap is available): ./venv/bin/python -m pytest -q tests/test_tool_path_confinement.py tests/test_workspace_confine.py tests/test_worker_runtime.py tests/test_containment_enforcement.py — all gated tests still run (no new skips) and pass, because enforcing_containment_available() probes the real mechanism table.
2. On a stock macOS host with the default /var/folders/wk/88p0jv_51m7d3x224pz1rsfh0000gn/T/: the same command — the previously-failing tests now pass or skip with the containment reason instead of erroring.
3. Full suite on macOS before/after: 81 failed, 14116 passed → 14185 passed, 96 skipped, 0 failed (CHROMADB_PORT=<closed-port> ./venv/bin/python -m pytest -q, ~25 min, with npm ci + npx playwright install chromium).
4. Confirm no production code changed: git diff dev...fix/macos-test-runtime --stat touches only tests/.