Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only test-case design skill whose sensitive topics are framed as QA test scenarios, not as hidden execution or data-access behavior.
Installers should treat this as a test-case-writing aid, not an execution harness. Review generated cases before running them, replace example passwords and bearer tokens with test-only placeholders, avoid production data or payment rails, and narrow prompts when asking for AI-agent tests because the "Agent" trigger is broad.
Using the standalone keyword 'Agent' as a trigger is highly ambiguous because it appears in many unrelated requests, including references to the assistant itself or generic business agents. This can misroute a wide range of prompts into agent-testing behavior, causing inappropriate file loads and instruction application in contexts the user did not intend.
The instruction mapping table repeats the ambiguous 'Agent' trigger, reinforcing the same unsafe routing logic in a second authoritative section of the skill. Duplicating the ambiguity increases the chance that implementations or maintainers preserve the overbroad behavior and apply the agent-testing path to unrelated prompts.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
- [ ] Important information has haptic feedback
## Storage & Data (10 items)
- [ ] SharedPreferences/Keychain read/write works correctly
- [ ] SQLite/Room/CoreData database operations work correctly
- [ ] Image caching strategy takes effect
- [ ] API data caching strategy takes effect
YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).
# AI Agent Testing
> This document defines specialized testing dimensions and strategy highlights for AI Agents (intelligent agents). Agents are cross-platform capabilities that can be embedded in Apps, Mini-Programs, H5, Desktop, PC Web, and other hosts, and are tested in conjunction with host platform testing.
> See `references/examples/format-spec.md` for output format and `references/checklists/agent-checklist.md` for the checklist.
---
## I. Task Completion
### Focus Areas
- User intent understanding accuracy: explicit instructions, implicit requirements, ambiguous requirements, intent
This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.
- Resource exhaustion handling: graceful degradation when memory/Token limits are reached
### Common Defects
- Being induced to bypass security restrictions and perform dangerous operations
- Leaking API Keys or internal information in logs/output
- Having role hijacked by user prompts
- Failing to decline clearly harmful requests
The skill description is very broad and can match many ordinary test-related requests without clear exclusions or activation boundaries. Overbroad routing can cause this skill to be invoked in unintended contexts, leading to irrelevant instructions being loaded and potentially displacing a more appropriate or safer skill.
This markdown file describes testing of authorizations, local storage, location, sharing, messaging, and payments, but it does not include any warning about potential effects on user privacy, stored data, or financial actions. For markdown files, omissions of warnings about behaviors that could affect user data, privacy, or system integrity should be flagged.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Leaking API Keys or internal information in logs/output
- Having role hijacked by user prompts
- Failing to decline clearly harmful requests
- Executing high-risk operations without confirmation
- Session data leakage when one instance serves multiple users
- Improper handling of extreme inputs (empty/overly long/special characters) leading to exceptions
- No reasonable degradation on service exceptions, resulting in direct failure
This markdown file includes example account credentials in a precondition and later shows an Authorization Bearer token in API test steps. Under the markdown variant of SQP-2, the document should warn users not to use real credentials or production secrets when following examples that affect privacy or system integrity.
The checklist includes a mandatory item for 'Chinese input method works correctly,' which imposes a specific language/locale requirement in natural language. Because the document does not state that it is intended for a China-specific product or offer locale scoping, this can be read as an unjustified language policy constraint.
The AI Agent testing example instructs creation of a file named test.txt, which is a write operation affecting system state. The markdown description does not include any warning that such steps should be confined to a safe test workspace or may modify existing files.
The testing guidance hard-codes shortcut notation like "Ctrl+C/V/Z/Y/A/S/F" and later references English UI text such as "Open with XX". In a natural-language/platform guidance document, this can amount to an implicit language/locale assumption without noting localization or user-language variation.
The guidance explicitly requires testing Chinese input method composition events, which imposes a language/locale-specific requirement in a general mobile web testing document. The file does not state that this checklist is limited to Chinese-language products or markets, nor does it offer locale-based applicability guidance.
The example says to paste a '10,000-character Chinese text,' which imposes a specific language in natural-language guidance. Under the policy, forcing a language or locale without user opt-in or a documented justification is a violation.
Detected: suspicious.exposed_secret_literal