systematic-debugging
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The find-polluter.sh script executes shell commands to run local tests via npm test to isolate state pollution issues.
- [PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze external data such as error messages and stack traces, creating a surface for indirect prompt injection. 1. Ingestion points: SKILL.md (Phase 1, Step 1) and root-cause-tracing.md (Section: Adding Stack Traces). 2. Boundary markers: Absent; no instructions are provided to distinguish data from executable commands in logs. 3. Capability inventory: Subprocess execution (testing and debugging scripts) and file system writes (diagnostic instrumentation). 4. Sanitization: Absent; the skill does not prescribe filtering or escaping of analyzed log content.
Audit Metadata