The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external financial data (tables) as its primary input. This creates a surface for indirect prompt injection if the user-provided data contains malicious instructions that could be misinterpreted by the agent during context construction.
Ingestion points: scripts/run.mjs reads user-provided files or text via the --input flag.
Boundary markers: The instructions do not define explicit delimiters to encapsulate or isolate the untrusted input from the agent's system prompt.
Capability inventory: The skill is configured with Bash for running the audit script and Read for file access.
Sanitization: While the internal engine performs numeric parsing, there is no explicit sanitization of the input text for prompt safety before it enters the agent's context.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for the agent to install the @alipay/agent-payment package via npx to facilitate a purchase workflow. This targets an official, well-known service from Ant Group.
[COMMAND_EXECUTION]: The skill relies on executing local Node.js scripts via the command line to perform its core auditing functions. This is the intended behavior of the tool.

subcontract-settlement-check-free