Open and answer this issue on GitHub.
ready · opened 2026-11-01 · updated 2026-11-01 · 0 comments
On 2026-11-01 at 23:32Z, K4 (k4-keeps-infer.md) and S1 (s1-sysroot-package.md) were each started a second time while their first run was still working. agent.nika saw <brief>.session and resumed the same session id, using the “was interrupted” prompt.
The two copies raced in the shared tree. The resumed copy overwrote keeps.nika and sysroot_package.nika, and lower-std then refused the doubled file with NK1148. Both first runs stopped without a report, at about $1.4 together; the debriefs k4-keeps-infer.debrief.md and s1-sysroot-package.debrief.md describe what happened. The resumed copies went on alone.
agent.nika writes its pid into <brief>.session beside the session id./proc/<pid>). If it is, it refuses with ESCALATE already running <brief> pid N and starts nothing.builder.md: a lower-std run takes run_in_background, because it takes longer than the 120 s default (both debriefs).