The Agent Skills Directory

[Prompt Injection] (SAFE): No instructions attempting to override agent behavior or bypass safety guidelines were found.
[Data Exposure & Exfiltration] (SAFE): No hardcoded credentials or access to sensitive local files (e.g., SSH keys, AWS configs) were identified. Network requests in examples are limited to localhost.
[Remote Code Execution] (SAFE): The skill does not perform remote script execution or download untrusted packages. It references standard local npm scripts for testing.
[Obfuscation] (SAFE): No encoded strings, hidden characters, or homoglyphs were found in the provided files.
[Indirect Prompt Injection] (LOW):
Ingestion points: Test examples in SKILL.md process JSON responses from /.well-known/webfinger.
Boundary markers: Absent in code snippets.
Capability inventory: The skill describes using npm run to execute test suites which involves subprocess calls.
Sanitization: None shown in the documentation snippets, but the risk is localized to the test environment.

activitypub-testing