The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external sources, including APK scan reports and live output from the Android Debug Bridge (adb). This data, which includes component names, provider authorities, and crash logs, is interpolated into the agent's context. While the tool categorizes results, the ingested content is not explicitly sanitized to remove potential instructions that could influence the agent's behavior.
Ingestion points: The load_inventory function in scripts/intentprobe.py reads metadata from apkscan.json, and the Device.shell method captures results and logs from the connected device.
Boundary markers: Findings are structured into specific result categories, providing some structural separation, but the skill does not use explicit delimiters or warnings to instruct the agent to ignore instructions within the raw data snippets.
Capability inventory: The skill has the ability to execute shell commands via adb and write JSON reports to the local filesystem.
Sanitization: The script employs shlex.quote to protect the target device from command injection, but does not sanitize the data for prompt injection patterns before presenting it to the agent.
[COMMAND_EXECUTION]: The Python script executes shell commands using the subprocess module to interact with the Android Debug Bridge (adb) utility. It invokes commands such as adb devices, adb shell, and adb logcat to probe application components.
Evidence: Found in the Device class and find_adb function within scripts/intentprobe.py.
Mitigations: The script correctly uses shlex.quote to escape arguments before they are passed to the device shell, which prevents command injection from malicious strings found in the application manifest.

android-intent-probe