construction-wage-special-account-check-free
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEPROMPT_INJECTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation in SKILL.md provides a specific text block for the user to copy and paste to the AI agent. This block contains instructions like "I have authorized the installation of dependencies and initiation of purchases" and "do not confirm for me" (不要替我确认). This is an attempt to bypass agent interaction guardrails by pre-asserting authorization for high-risk actions.\n- [REMOTE_CODE_EXECUTION]: The documentation suggests the agent execute 'npx -y @alipay/agent-payment@latest install'. This command downloads and executes code from a remote registry at runtime. While the target is a well-known service (Alipay), the runtime execution of remote code initiated by a prompt injection payload is a significant capability that bypasses standard installation reviews.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary input data representing wage tables which could contain malicious instructions. 1. Ingestion points: Untrusted data enters via 'payload.text' in scripts/engine/construction-wage-special-account-check.js. 2. Boundary markers: Absent; there are no delimiters instructing the agent to ignore embedded commands. 3. Capability inventory: Use of the Bash tool is permitted in SKILL.md, and scripts/run.mjs has access to the filesystem via the fs module. 4. Sanitization: Absent; the skill uses regular expressions for splitting and basic number conversion but does not escape or sanitize content against prompt injection.
Audit Metadata