error-diagnostics-error-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as error reports, logs, and traces through the
$ARGUMENTSvariable without using boundary markers or specific instructions to ignore embedded commands. This creates a surface where malicious instructions hidden in a stack trace or log file could influence the agent during the investigation or reproduction steps. - Ingestion points: The
$ARGUMENTSfield inSKILL.mdallows the entry of arbitrary error data. - Boundary markers: There are no explicit delimiters or warnings to treat the content as data only.
- Capability inventory: The skill allows the agent to 'reproduce or narrow the issue with targeted experiments' and 'propose fixes,' which involves command execution and file modification.
- Sanitization: No sanitization or filtering of external content is described.
Audit Metadata