The Agent Skills Directory

[PROMPT_INJECTION]: Deceptive authorship metadata. The skill's frontmatter incorrectly claims the author is 'Google LLC', whereas the actual provider is 'albertmartorell1975'. This impersonation misleads users about the official status and security validation of the skill's instructions.\n- [EXTERNAL_DOWNLOADS]: Risky external file acquisition. The documentation (in references/android/identity/digital-credentials/index.md) directs users to download and install a pre-compiled Android APK (app-debug.apk) from the GitHub Actions artifacts of an unverified third-party repository (digitalcredentialsdev/CMWallet). Installing executable files from untrusted sources is a critical security risk.\n- [REMOTE_CODE_EXECUTION]: Manual execution of untrusted binary code. By instructing users to install an APK from a non-official source, the skill facilitates the execution of unvetted code on the user's device.\n- [DATA_EXFILTRATION]: Potential PII exposure via system logs. The implementation examples suggest using Log.d to output verified user details (email and name) to the Android system log. This can leak sensitive identity information to any other application on the device with permission to read system logs.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill describes a workflow that ingests and processes untrusted digital credentials from external sources.\n
Ingestion points: The skill reads external JSON data via the Credential Manager API response (credential.credentialJson).\n
Boundary markers: While the documentation warns about the need for server-side validation, there are no programmatic boundaries provided in the snippets.\n
Capability inventory: The skill provides Android code templates but does not grant the agent direct access to sensitive system capabilities based on the ingested data.\n
Sanitization: The instructions recommend server-side cryptographic verification and the use of nonces to mitigate risks.

verified-email