stress-strategy

name	stress-strategy
description	Build multi-profile stress tests from brute complexity and constraints, then execute with traceable evidence.
disable-model-invocation	false

Stress Strategy Skill

Goal: build a reproducible and explainable multi-profile stress strategy from brute complexity and constraints, instead of running a single random profile.

Rules

• Include at least 3 profiles: tiny_exhaustive, random_small, and edge_small.
• Use type=1 for tiny_exhaustive, type=2 for random_small, and type=3/4 for edge_small.
• Keep semantic boundaries explicit:
  • type=3: boundary and extreme constraints coverage.
  • type=4: targeted worst-case/TLE-inducing patterns, not simple max-parameter scaling.
• When brute complexity is higher than O(n^2), automatically lower n_max and trials.
• Every profile must be reproducible (fixed seed rules and replayable parameters).

Execution

When calling stress_test_run, pass stress_profiles and record completed rounds and failure points for each profile.

Output Requirements

• decision: go / no_go
• stress_profiles: each profile's name/trials/types/generator_args
• execution_summary: completion status and failed rounds per profile
• next_fix_hint: if failed, provide prioritized fix direction (std/brute/generator/validator)

Decision Rules

• go: all required profiles complete and no correctness mismatch remains.
• no_go: any required profile is incomplete, or mismatches remain unresolved.

Forbidden Behavior

• Do not run only a single random profile.
• Do not treat type=4 as a duplicate of type=3.
• Do not continue after unresolved mismatches.