AGENT SKILL

torture-test

Scaffold endurance, stress, and adversarial-input tests that reveal degradation rather than immediate crashes

License 许可
MIT
所属仓库
jmagly/aiwg
最近更新
2026-11-14
Stars
★ 229
Forks
36

01 前言 Front Matter

name
torture-test
errors
[ { "no-target": "no candidate surface detected and no explicit target supplied" } ]
ensures
[ { "scenarios": "endurance, stress, and adversarial-input scenario templates" }, { "ci-recipe": "long-form nightly or scheduled test recipe distinct from PR-gating" } ]
requires
[ { "target": "parser, network service, allocator, CLI stdin handler, or other untrusted-input surface" } ]
namespace
aiwg
platforms
[ "all" ]
invariants
[ "torture tests are long-form jobs, not unbounded PR blockers", "pass/fail criteria include resource ceiling and recovery behavior" ]
commandHint
{ "model": "haiku", "category": "testing", "modelRole": "efficiency", "modelTier": "economy", "allowedTools": "Read, Write, Bash, Grep", "argumentHint": "[--target <path-or-service>] [--type endurance|stress|adversarial|all] [--ci github|gitea|gitlab]", "orchestration": false }
description
Scaffold endurance, stress, and adversarial-input tests that reveal degradation rather than immediate crashes

03 SKILL.md 正文

Torture Test

Design long-form torture tests for high-load, malformed-input, timing, and resource-exhaustion scenarios. This complements fuzzing-in-ci: fuzzing finds inputs that crash; torture testing finds inputs that degrade, leak, starve, or recover incorrectly.

Scenario Types

  • Endurance: repeated valid workloads over long duration with memory/FD/latency ceilings.
  • Stress: high concurrency, burst input, large payloads, queue pressure.
  • Adversarial input: slow-loris timing, malformed grammar sweeps, partial reads, truncation, retry storms.

Outputs

The skill emits scenario cards under .aiwg/test/torture/ and scheduled CI recipes under .aiwg/test/torture/ci/. PR-gating remains short; torture suites run nightly, hourly, or on demand.

Pass/Fail Criteria

Every scenario defines memory ceiling, file descriptor ceiling, latency tail, error recovery time, and acceptable error classes.

References

  • agentic/code/frameworks/security-engineering/skills/fuzzing-in-ci/SKILL.md
  • .aiwg/security/curl-checklist-gap-analysis.md row 2, Practice 11

04 同仓库更多 Skills