diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill methodology requires the agent to ingest and analyze untrusted external data (such as bug reports, logs, HAR network traces, and core dumps) and then use that data to drive feedback loops involving shell commands and browser automation. This creates a surface for indirect prompt injection where malicious instructions embedded in data could affect the agent's behavior.\n
- Ingestion points: Processes repository files (CONTEXT.md, ADRs), log dumps, network request captures (HAR files), and system core dumps.\n
- Boundary markers: The instructions do not provide specific delimiters or warnings to isolate untrusted data from the agent's control flow.\n
- Capability inventory: Encourages shell command execution, curl usage, test suite triggers, and headless browser scripts (Playwright, Puppeteer).\n
- Sanitization: There are no specified procedures for validating or sanitizing the external data before it influences the debugging workflow.
Audit Metadata