The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Attack Surface
Ingestion points: The skill accepts untrusted Office documents (e.g., .docm, .xlsm) as input via the target parameter in scripts/agent.py and the manual workflows described in SKILL.md.
Boundary markers: The skill's output (reports and extracted code) does not implement delimiters or explicit 'ignore instructions' markers when presenting extracted macro content to the AI agent.
Capability inventory: The skill performs VBA extraction, deobfuscation, and metadata analysis across scripts/agent.py and SKILL.md.
Sanitization: There is no evidence of sanitization or escaping of the extracted VBA code or document metadata, which may contain malicious instructions targeting the AI model.
[EXTERNAL_DOWNLOADS]: Installation of Third-Party Dependencies
The skill requires several external Python packages including oletools and xlmdeobfuscator, as noted in SKILL.md and references/api-reference.md. These are widely recognized industry-standard tools for malware analysis.
References a download for oledump.py from Didier Stevens' well-known security blog.
[COMMAND_EXECUTION]: Shell Command Execution for Malware Analysis
SKILL.md and references/api-reference.md provide numerous examples of shell commands (e.g., olevba, oledump.py, xlmdeobfuscator) intended for use in an analysis environment. While these are legitimate tools, executing them on untrusted files carries inherent risks if the analysis environment is not properly isolated.

analyzing-macro-malware-in-office-documents