daily-digest-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFECREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted content from external memory files.\n
- Ingestion points: The
digest_daily.pyscript reads markdown files from thememory/directory.\n - Boundary markers: The
SKILL.mdfile contains instructions to treat memory content as data rather than instructions, providing a conceptual boundary.\n - Capability inventory: The skill has the capability to read from
memory/and write tojournals/digest/using standard file system operations.\n - Sanitization: The Python script performs no sanitization or validation of the input data before generating the digest output.\n- [CREDENTIALS_UNSAFE]: The
SAFETY.mdfile contains a hardcoded example credentialddapi_verysecret123456789abcwithin a safety evaluation test case. While used for demonstration purposes in a safety context, hardcoding strings that match realistic credential patterns in the skill distribution is a security best-practice violation.
Audit Metadata