The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted application source code and manifest files provided by the user. This data is scanned by scanner.py and processed by orchestrator.py to identify policy-related signals. A malicious application could contain comments, string literals, or metadata designed to mislead the auditor or inject instructions into the generated compliance reports.
Ingestion Points: The audit flow recursively walks and reads all source, configuration, and manifest files from the target app_dir.
Boundary Markers: The skill lacks explicit boundary markers or instructions for the agent to ignore embedded prompts within the analyzed application files.
Capability Inventory: The skill is capable of executing local Python scripts, performing file system operations within the workspace, and accessing Google Play Store domains via network requests.
Sanitization: Sanitization is limited to emoji stripping in the final report generator, leaving a surface for malicious content to influence the agent's interpretation.
[DYNAMIC_EXECUTION]: The orchestrator.py script utilizes a template engine to dynamically generate markdown instruction files (prompt_worker_*.md) and data payloads (input_worker_*.json). These instructions are assembled at runtime based on the signals detected during the initial codebase scan and are then used to direct the behavior of subsequent agent turns or sub-agents.
[EXTERNAL_DOWNLOADS]: The skill employs play_store_scraper.py to retrieve application metadata, store descriptions, and data safety declarations from the Google Play Store (play.google.com). This cross-referencing is essential for the audit process. The target domain is a well-known and trusted service for this application context.
[COMMAND_EXECUTION]: The SKILL.md file instructs the agent to execute specific local Python scripts, including orchestrator.py and generate_report.py. These scripts manage the core logic of the audit lifecycle, including environment initialization, findings aggregation, and report rendering.

play-policy-insights