The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted mobile application files (APK, IPA, AAB) and decompiled source code. An adversary could potentially craft a malicious application containing specific strings or metadata designed to influence the agent's behavior during the analysis phase.
Ingestion points: Processes external application packages and resources via functions like inspect_android_package, decode_android_resources, and identify_flutter_build in the REA MCP toolchain.
Boundary markers: No specific delimiters or "ignore embedded instruction" prompts are defined to separate tool output or decompiled strings from the agent's primary instructions.
Capability inventory: The skill uses powerful tools for file system manipulation (Apktool, 7-Zip), dynamic runtime instrumentation (Frida), and binary analysis (Ghidra, REA MCP), which provide a significant capability set if an injection were successful.
Sanitization: The instructions do not specify sanitization or filtering of the extracted code or manifest content before the agent evaluates it for vulnerabilities.
[COMMAND_EXECUTION]: The toolchain requires the use of numerous command-line utilities (e.g., adb, apktool, frida, sqlite3, aapt2) to interact with the target environment and binary files. These operations are essential for the primary purpose of mobile security auditing but involve direct interaction with external tools and processes.

rea-mobile-app-recovery