qa-security-sweep
Adversarial QA audit that turns security findings for a feature, module, or diff into canonical Layer:security regression scenarios and archived findings.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 13
- Forks
- 6
01 前言 Front Matter
- name
- qa-security-sweep
- description
- Adversarial QA audit that turns security findings for a feature, module, or diff into canonical Layer:security regression scenarios and archived findings.
- allowed-tools
- Read, Write, Grep, Glob, Bash, ApplyPatch
- argument-hint
- [<feature-slug>|--diff <ref>|--pr <number>] [--write] [--severity-floor critical|high|medium|low]
- disable-model-invocation
- false
02 资源文件 2
- 参考文档commands/qa-security-sweep.md2.2 KB
- 参考文档manifest.yaml841 B
03 SKILL.md 正文
QA Security Sweep
Purpose
Treat security testing as a first-class QA layer. The skill reviews a feature,
module, or diff with an adversarial mindset, then emits findings as canonical
Layer: security regression scenarios that can be tracked by qa drift
and implemented as browser/API/security tests.
Template paths are resolved from ${AGENTS_SKILLS_ROOT}/qa-security-sweep/.
Scope
Inspect every relevant layer in the target repo:
- frontend UI, client state, browser storage, forms, URLs, iframes, messages
- backend route handlers, controllers, server actions, jobs, webhooks
- authn/authz, sessions, role checks, tenant isolation, policy enforcement
- database queries, migrations, RLS/policies, secrets, audit logs
- integrations, webhooks, credential handling, replay protection
- infrastructure headers/config, environment handling, CI/CD, dependencies
- business logic, state machines, race conditions, workflow bypasses
Steps
- Follow
${AGENTS_SKILLS_ROOT}/qa-security-sweep/commands/qa-security-sweep.md. - Load the QA profile and feature catalog when present.
- Identify existing scenarios and tests for the target feature/prefix.
- Build a threat model before writing scenarios.
- Propose
Layer: securityscenarios using canonical IDs and the sharedSecurity Class:vocabulary. - Archive the full findings report under
qa/security/findings/when the repo uses that convention, or under the QA profile's configured security output path when present. - With
--write, append scenarios to the relevant regression spec files and runqa drift.
Guardrails
- Do not actively exploit live infrastructure.
- Do not include real secrets, tokens, customer data, or production identifiers.
- Prefer testable regression scenarios over vague prose findings.
- Model exploit chains when individually low-severity issues compose into a higher-risk path.
- Use repo-local fixture IDs and helper patterns when proposing tests.
04 同仓库更多 Skills
personal-knowledge
Save sources to the private Harnessy learning library, sync delivered readings, answer questions from saved evidence, browse its wiki, and generate weekly learning reviews.
valid-skill
A valid skill for testing purposes.
unknown-field
A skill with an unknown frontmatter field.
nested-child
Nested skill should be ignored.
root-skill-preferred
Root skill should win.
no-frontmatter
暂无简介