The Agent Skills Directory

[SAFE]: The skill instructions define professional QA workflows and methodologies, such as the Testing Pyramid and AAA (Arrange-Act-Assert) pattern, which are standard industry practices.
[DATA_EXPOSURE_AND_EXFILTRATION]: Analysis of the instructions and configuration files revealed no hardcoded credentials, API keys, or suspicious file access patterns targeting sensitive system directories.
[REMOTE_CODE_EXECUTION]: There are no commands or instructions that attempt to download and execute untrusted remote scripts or install unverified packages from third-party sources.
[PROMPT_INJECTION]: The skill does not contain any phrases or structures intended to override agent safety protocols, ignore previous instructions, or extract system prompts.
[INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The agent is designed to analyze user-provided software requirements and source code to generate tests (SKILL.md).
- Boundary markers: Explicit delimiters for untrusted data are not provided in the prompt templates.
- Capability inventory: The configuration for Claude Code requests broad permissions including bash, read, write, and webfetch, which are necessary for testing but represent a capability tier that could be exploited if malicious instructions are present in the ingested code.
- Sanitization: The skill does not define specific sanitization routines for external data.
- Note: This is an inherent design characteristic of a QA agent and does not indicate malicious intent.

qa-engineer-agent