error-diagnostics-error-analysis
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Potential surface for indirect prompt injection discovered in the data processing flow.\n- Ingestion points: The skill ingests untrusted data in the form of error messages, stack traces, and logs via the
$ARGUMENTSplaceholder inSKILL.md.\n- Boundary markers: There are no instructions to use delimiters or ignore embedded commands within the provided error context.\n- Capability inventory: The agent is instructed to reproduce issues and propose fixes, which are capabilities that could be exploited if malicious instructions are embedded in the logs.\n- Sanitization: The instructions lack requirements for sanitizing or filtering the content of the diagnostic data before processing.
Audit Metadata