The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill correctly identifies that reviewed apps are a source of untrusted data. It includes a mandatory safety policy in SKILL.md instructing the agent to treat UI text and logs as data only and never as instructions, effectively mitigating potential injection attacks from the application under test.
[SAFE]: The scripts/report.py utility employs strict HTML escaping for all externally sourced data (app labels, reviews, hostnames, crash logs) before rendering them into the final report. This prevents XSS or other injection vulnerabilities when the report is viewed in a browser.
[SAFE]: The skill provides clear guidance on credential safety, instructing users to use disposable emulators and throwaway accounts, and explicitly forbidding the entry of real credentials or payment methods.
[SAFE]: The prompt injection warning identified by static detectors in SKILL.md is a false positive. The phrase "ignore previous instructions" appears as an example of malicious data the agent should watch out for and ignore, rather than an attempt to bypass the agent's own safety filters.

android-app-review