error-debugging-error-analysis
Pass
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to the ingestion of untrusted external data through the $ARGUMENTS parameter.
- Ingestion points: The skill accepts error context, stack traces, and log files in SKILL.md, which may be attacker-controlled.
- Boundary markers: There are no explicit delimiters or instructions to ignore commands within the data, which may lead the agent to follow instructions embedded in logs.
- Capability inventory: The agent is authorized to conduct targeted experiments and propose code fixes based on the analyzed data.
- Sanitization: While the skill suggests redacting PII, it lacks mechanisms to sanitize or validate input for malicious instructions.
Audit Metadata