diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the codebase, project documentation (CONTEXT.md, ADRs), and user-provided bug reports to guide its diagnosis process. It also has the capability to execute shell commands, network requests, and test suites. Malicious instructions embedded in these files or reports could attempt to manipulate the agent's behavior during the diagnosis loop.
- Ingestion points: Codebase modules,
CONTEXT.md, ADRs, and user-provided symptoms or artifacts. - Boundary markers: No specific delimiters or instructions to ignore embedded commands are mandated when reading project files.
- Capability inventory: The agent is authorized to run
curl,git bisect, Playwright/Puppeteer, and arbitrary reproduction scripts. - Sanitization: No sanitization or validation of the content read from the codebase is specified.
- [COMMAND_EXECUTION]: The skill's core functionality relies on the agent executing various command-line tools and scripts to reproduce and minimize bugs. This includes using
curl,git, and testing frameworks as part of the Phase 1 feedback loop. - [EXTERNAL_DOWNLOADS]: The skill suggests using tools like Playwright and Puppeteer for headless browser scripts. These are well-known industry-standard developer libraries that may download browser binaries during setup.
Audit Metadata