diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data sources to reproduce bugs. Maliciously crafted artifacts could attempt to influence agent behavior.
- Ingestion points: Phase 1 instructions specify reading log dumps, HAR files, and captured traces (SKILL.md).
- Boundary markers: No specific delimiters or "ignore instructions" markers are defined for the analysis of external artifacts.
- Capability inventory: The skill utilizes shell command execution, file system access, and network operations (curl) to build reproduction loops (SKILL.md).
- Sanitization: The skill provides instructions for the agent to redact its own output (e.g.,
<REDACTED>), but lacks sanitization or validation rules for incoming external content. - [DYNAMIC_EXECUTION]: The skill's methodology involves the agent dynamically creating and executing local scripts.
- Evidence: Phase 1 and Phase 6 describe using a shell script template (
scripts/hitl-loop.template.sh) which the agent is expected to customize and execute to drive human-in-the-loop interactions. - [COMMAND_EXECUTION]: The agent is granted a broad set of execution capabilities to fulfill the diagnostic purpose.
- Evidence: The skill instructs the agent to run tests, curl/HTTP scripts, CLI invocations, and headless browser scripts (Playwright/Puppeteer) to establish a feedback loop (SKILL.md Phase 1).
Audit Metadata