[brujack]

Summary

• The Test job ran three full builds and test runs one after another: nextest (288 s), the tarpaulin gate (327 s) and the informational network tarpaulin (325 s). That is about 16.5 min on the merge path (PR fix(claude): update each listed plugin once #156, run 41138976453).
• Each tarpaulin run is now its own job, so it runs beside test on a separate runner:
  • coverage: the 81% gate and the badge push. It is in auto-merge needs:.
  • coverage-network: informational. It has job-level continue-on-error: true and is not in needs:. The ALLOWED entry still matches by name; its reason is updated.
• test and coverage-network drop to contents: read. Only coverage keeps contents: write, for the existing coverage-data badge push.
• CLAUDE.md's CI table and "Which jobs block" paragraph are updated. The "Run coverage as its own job" backlog row is removed.
• Coverage was added to main's required status checks after this run reported it (operator-approved). Branch protection is as strong as when tarpaulin ran inside Test.

Test Plan

• pytest tests/: 167 passed, including tests/test_continue_on_error.py, which still matches the job-level site.
• The YAML parses. The structure was checked: needs:, permissions, and job-level continue-on-error on coverage-network only.
• No Rust test needs shellcheck, ruff or pytest, so the coverage jobs install Rust and tarpaulin only.
• This PR's own run (41149333695): Test took 4m34s, Coverage 5m34s and Coverage with network tests (informational) 6m48s, all in parallel. The PR merged 6 min after it opened; before the split, Test alone took about 16.5 min.
• Coverage (run 41149333695): Rust 83.55% (5283/6323, floor 81); Python 95.28% (floor 90).

Code review

reviewed-sha: 8bc50f2
diff-hash: 6d59c3dc979c1fc2fd401db3c20842ff99a9d95aa013e991a9f9265121a2d5f9
reviewers: 5/5

Three rounds. Round 1 (0f58d46): two findings, fixed in 1c29943 (stale ALLOWED reason "the main coverage step gates"; moved step comment said an 80% gate against --fail-under 81). Round 2 (1c29943, verification): one finding, fixed in 8bc50f2 (comment said "the three" tarpaulin runs; there are two, after nextest). This was past the two-round fix cap; the operator chose to fix it and run one more check. Round 3 (8bc50f2, verification): no findings. Steps 2-3 ran inline: root CLAUDE.md is the only one in scope.

RESOLVED: 3 fixed, 0 kept

Spec alignment

SKIP: no open plan has a spec: line (backlog row, no plan)

🤖 Generated with Claude Code

https://claude.ai/code/session_01RMRWKLc4oBDsXjYryqoNxf