Round 8: all eight real, fixed in e67f95d. Two of them contradicted each other, which was the interesting part.
ed177217asked me to extend the engine-spec exemption to a bareopencode;…
Round 7: 13 of 14 real, fixed in 3af0f09. Seven of them, from all four models, are the same finding — and it's the worst one in this PR.
**🔒 I fixed the cross-vendor credential leak on…
Round 6: all five real, fixed in a3d3a45 + 274451e. Two were security-relevant, and chasing one of them uncovered something no reviewer mentioned.
**The majordomo bump this PR depends on was…
Round 5: all ten real, fixed in 0abcd16. The best findings are again about the previous round's fix — three separate holes in my credential scrub, each of which would have made it useless in…
Round 4: all six real, fixed in 14f8533. One of them is the most valuable finding of this PR and both Claude models found it independently.
**🔒 The test job I added leaked a push-capable…
Round 3: all twelve real, fixed in 67a7361. The two that matter are about the pre-flight I added, not about qwen.
run.sh:197— the false pass in the other direction. Round 2 caught…