Reflects inter-task data chaining via ScratchBuffer, all 84 tasks consuming
work data, new COM/WMI category with 8 tasks, updated anti-detection table,
corrected test counts (185 verified), and added cat-com to feature flags table.
Tasks can now accept arbitrary user variables via .feed() on the builder.
Data is cloned on entry and woven into task control flow (XOR into buffers,
seed derivation for loop bounds, hash inputs, cipher keys), making busywork
blocks indistinguishable from real data processing under data-flow analysis.
31 tasks across COMPUTE, MEMORY, and CRYPTO actively consume fed data.
All originals remain untouched — tasks receive read-only references to clones.
Test suite covers: per-type serialization (52), direct invocation of every
task with 10 work-data shapes x parameter extremes (18), mirror tests proving
data actually changes computation (28), data isolation (15), stress/regression
(23), and integration across all categories, intensities, and edge cases.
Explains the EDR/anti-cheat evasion purpose, why sleep() is a detection
signal, and how each design decision maps to a specific detection
technique it defeats. Includes example trace showing varied syscall
sequences across consecutive calls.