office-document-specialist-suite-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains adversarial prompt examples in
SAFETY.md(e.g., 'ignore previous instructions') which are used for training and demonstrating defensive guardrails rather than for malicious intent. It also addresses the surface for indirect prompt injection by mandating explicit file specification and absolute path resolution inSKILL.md. Ingestion points: Document parsing inods.py. Boundary markers: Explicit instructions inSKILL.md. Capability inventory: File system writes inods.py, dependency installation insetup.sh. Sanitization: Content validation is deferred to the agent through guardrails. - [COMMAND_EXECUTION]: The
setup.shscript executes shell commands to initialize a Python virtual environment and install declared dependencies from PyPI, following standard development practices. - [DATA_EXFILTRATION]: The skill identifies sensitive paths such as
~/.ssh/and.envin its documentation to provide the agent with negative constraints, explicitly prohibiting unauthorized access to credential files.
Audit Metadata