fix(skills): check codecov comment before local coverage build

[worktrunk-bot]

Summary

The bot repeatedly installs cargo-llvm-cov from scratch in CI review sessions
to investigate codecov/patch failures. This wastes significant compute time
(~5-10 min per occurrence). The fix guides the bot to check codecov's PR comment
first — most failures can be assessed as false positives without local coverage.

Evidence: 4 occurrences across sessions (3 historical + 1 this hour in run
23424793547 reviewing PR #1683). The bot installed cargo-llvm-cov, ran the full
test suite with coverage twice (PR branch + main), consuming ~10 minutes of
compute to conclude the failure was a false positive in #[cfg(unix)] TUI code.

Root cause: The running-in-ci skill's codecov investigation step jumped
directly to task coverage / cargo llvm-cov report, which requires
cargo-llvm-cov to be installed. The skill lacked guidance to check the readily
available codecov PR comment first.

Fix: Replace the direct task coverage instruction with a triage step:
check codecov's PR comment to identify uncovered files/lines, and only run local
coverage when the uncovered lines are in core testable logic.

Gate assessment:

• Evidence level: High (4 occurrences, consistent pattern)
• Change type: Targeted fix (replaces 2 lines with 7 lines of guidance)
• Historical evidence: review-reviewers tracking review-reviewers tracking: 2026-03 #1611

Test plan

• Verify next codecov/patch failure investigation checks PR comment first
• Confirm cargo-llvm-cov is only installed when needed for core logic gaps

🤖 Generated with Claude Code

[max-sixty]

Superseded by tend extraction (#1676). Target file (.claude/skills/running-in-ci/SKILL.md) no longer exists. The codecov comment check can be added to the running-tend overlay if needed.

[worktrunk-bot]

Understood — makes sense that the tend extraction superseded this. I'll add the codecov-comment-first triage step to the running-tend overlay if the pattern recurs.