The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation includes a specific instruction block intended for the AI agent that directs it to skip user confirmation dialogs. Specifically, it uses the instruction '䐂要晡我确认' (do not confirm for me) during software installation and purchase flows, attempting to override the agent's interaction-policy and safety-confirmation constraints.
[REMOTE_CODE_EXECUTION]: The skill references the execution of npx -y @alipay/agent-payment@latest install for its purchase workflow. This command downloads and executes code from the Alipay organization's official registry. As Alipay is a well-known service provider, this is documented neutrally as a functional external dependency reference.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external financial data tables, creating a potential surface for instructions embedded in data.
Ingestion points: External text data is ingested through the --input file flag or the JSON text property in scripts/run.mjs.
Boundary markers: The implementation lacks delimiters or instruction-isolation markers to distinguish between data content and agent instructions.
Capability inventory: The skill uses the Read and Bash tools, performing local file system reads via fs.readFileSync.
Sanitization: While numeric fields are normalized, the text-based fields are processed as raw strings without escaping or sanitization.

asset-impairment-check-free