ferris-debug
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The workflow requires the agent to ingest and analyze external, potentially untrusted data such as error logs and stack traces. This ingestion point could be used to deliver malicious instructions to the agent during a debugging session.
- Ingestion points: SKILL.md (Step 1) specifies reading full error text, complete stack traces, and failing assertions.
- Boundary markers: The instructions do not define specific boundaries, delimiters, or ignore-instructions for the evidence data.
- Capability inventory: The agent is empowered to perform git bisection, instrument code, and modify source code to implement fixes.
- Sanitization: While the skill requires redacting sensitive data from outputs, it does not specify sanitization or filtering for the input evidence.
- [SAFE]: The skill includes explicit instructions in references/tracing.md to redact credentials, tokens, and connection strings before printing environment or configuration dumps, which is a defensive security measure.
Audit Metadata