engineering-prompt-engineer
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains phrases like "Ignore all previous instructions" and "Ignore instructions". These are used as examples within a test suite template and documentation for injection defense strategies. They do not represent an attempt to override the agent's operational logic or bypass safety filters.
- [REMOTE_CODE_EXECUTION]: The skill provides Python code snippets for automated prompt testing and modular assembly. These snippets are presented as deliverables for the user's local environment and are not configured for automatic execution by the agent.
- [DATA_EXPOSURE]: There are no patterns suggesting the harvesting of sensitive files (e.g., credentials, SSH keys) or hardcoded secrets. File operations mentioned in templates are limited to reading local prompt markdown files.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing user requirements into LLM prompts, which is an inherent ingestion surface.
- Ingestion points: User-provided behavioral specifications and prompt requirements.
- Boundary markers: The instructions mandate the use of XML-style tags (
<thinking>,<answer>,<example>) and reasoning scaffolds to separate instructions from data. - Capability inventory: No file system writes, network requests, or shell commands are performed by the skill itself.
- Sanitization: The instructions include a dedicated "Prompt Injection Defense" capability, requiring the agent to build injection-resistance layers and content boundary checking into the prompts it generates.
Audit Metadata