test-qa
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to crawl and interact with live web applications during Phase 2 (Intelligent Page Crawl). This creates a surface for indirect prompt injection where malicious instructions embedded in the target website's DOM, metadata, or console output could influence the agent's behavior.
- Ingestion points: Browser snapshots, console logs, and network request data captured via
playwright-cli(SKILL.md, Phase 1 & 2). - Boundary markers: None present to delimit untrusted web content from agent instructions.
- Capability inventory: Shell command execution (
npx), file system discovery, and database query execution (supabase:execute_sql). - Sanitization: No explicit sanitization or filtering of external web content is mentioned before processing.
- [DATA_EXFILTRATION]: The skill instructions direct the agent to read sensitive configuration files to discover test credentials and environment details. While the instructions advise against pasting secrets directly, the agent still accesses these files.
- Evidence: Accesses
.env.local,.env.test, and.env.examplein Phase 0d (Discover Auth Pattern). - [EXTERNAL_DOWNLOADS]: The skill dynamically downloads and executes the Playwright CLI tool from the NPM registry at runtime.
- Evidence: Uses
npx --yes @playwright/cli@latestto initialize the browser environment (SKILL.md, Phase 1b). - [COMMAND_EXECUTION]: The skill executes shell commands for browser automation and database interactions.
- Evidence: Executes
npxcommands for Playwright andsupabase:execute_sqlfor data pipeline verification (SKILL.md, Phase 4f). - [DYNAMIC_EXECUTION]: The skill permits the use of
evalorrun-codefor inspection purposes. - Evidence: "
eval/run-codeinspection-only" (SKILL.md, Critical rules).
Audit Metadata