The Agent Skills Directory

[PROMPT_INJECTION]: Deceptive Metadata. The skill's metadata in SKILL.md claims the author is "Google LLC", whereas the platform identifies the author as "albertmartorell1975". This impersonation of a well-known entity is misleading and could cause users to misjudge the skill's official status and the reliability of its results.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill performs static analysis on untrusted Android application code and incorporates found code snippets (evidence) into prompts for sub-agents or its own future turns. This creates a surface where malicious comments or string literals in the analyzed app could attempt to influence the audit outcome or the agent's behavior.
Evidence: Found in scripts/orchestrator.py (lines 538-600) where base_context["data_sources"] is populated with code snippets and passed to write_agent_prompts, which then generates new instructions containing this untrusted data.
[COMMAND_EXECUTION]: Standard Operation. The skill relies on executing local Python scripts (orchestrator.py, generate_report.py) to perform its audit functions. While this is the intended purpose of the tool, it involves significant interaction with the local shell and file system.

play-policy-insights