| Field | \nBound contract | \n
|---|---|
| Trigger | \nBranchy hot code underperforms, a likely or branchless refactor needs judgment, a kernel mitigation such as retpoline or KPTI needs explaining, or code branches on secret data. | \n
| Authority | \nRead-only. The skill runs perf stat on a user-named binary, reads sysfs, and answers in chat. Nothing on disk changes, so there is nothing to roll back. No remote mutation. | \n
| Side effect | \nChat output only. perf stat writes counters to stdout. | \n
| Done | \nThe answer names the branch that mispredicts or the speculation path that leaks, gives a measured branch-misses count where a binary exists, and states the fix with the condition under which it helps. | \n
branch-misses count and its ratio to branches for the real workload is recorded, or no binary exists and the answer is marked as unmeasured.perf stat -e branches,branch-misses ./app\nRead the ratio against the workload, not against a fixed number: a tight loop over sorted data should show a ratio near zero, while a parser over random input can sit far higher and still be at its floor. Only a ratio that drops after a change proves the change.
\nint m = a < b ? a : b; may lower to cmov. The select executes both operands every time, so it wins only when the branch mispredicts often; on a predictable branch it loses.__builtin_expect(!!(x), 1) and __builtin_expect(!!(x), 0) steer code layout, not the dynamic predictor. On a current out-of-order core the predictor already learns most static patterns, so the hint helps layout of the cold path and little else. Measure after adding one. Done when: the hint is kept only with a measured win./sys/devices/system/cpu/vulnerabilities/ (one file per issue, such as spectre_v1, spectre_v2, meltdown, l1tf). Retpoline, IBRS, IBPB, and STIBP appear in the spectre_v2 text. A host booted with mitigations=off reports Vulnerable here.-mspeculative-load-hardening masks pointers on the speculative path.For the pipeline model behind the penalty, use cpu-pipelines-and-hazards. For cache timing channels, use cpu-cache-opt. For the kernel mitigation set (KPTI, CET), use kernel-security.
| Failure class | \nBehavior | \n
|---|---|
| No binary or workload | \nDeliver the mechanism and the candidate fixes as hypotheses. Mark the answer unmeasured. | \n
perf stat denied | \nReport the perf_event_paranoid value the tool prints and the capability it needs. Do not change the sysctl. | \n
| Hint shows no gain | \nThe predictor already handled the branch. Remove the hint and profile for the real bottleneck. | \n
| Branchless version slower | \nThe branch was predictable and the select now executes both operands. Revert and benchmark on the target CPU. | \n
| Mitigation regresses throughput | \nName the mitigation and its cost. Isolating the secret-handling code is the alternative; do not recommend disabling mitigations. | \n
A chat answer that names the mechanism, the measured branch-misses figure when a binary exists, one chosen remedy with its winning condition, and the mitigation layer that applies when secrets are involved.