The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains a block of text specifically designed to be copied and sent by the user to the AI assistant. This text instructs the agent to perform actions such as installing external packages via npx and initiating a purchase through SkillPay, which is a form of directive that overrides the agent's current task context.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data (such as employee names and table headers) and interpolates them directly into report messages (e.g., in the who and checkTaxable functions in scripts/engine/iit-withholding-check.js). If an input file contains malicious instructions in these fields, the agent might interpret the resulting report as new instructions.
Ingestion points: Data is read from user-provided files via the --input argument in scripts/run.mjs and processed by scripts/engine/iit-withholding-check.js.
Boundary markers: The skill does not implement delimiters or warnings to isolate user-provided data from the agent's reasoning process when it reads the findings.
Capability inventory: The skill environment is configured with Read and Bash permissions.
Sanitization: Input strings are used directly in message templates without escaping or validation against injection patterns.
[EXTERNAL_DOWNLOADS]: The skill's documentation references the download and installation of the @alipay/agent-payment package from the npm registry. This package is an official component of a well-known service used for managing agent-based transactions.

iit-withholding-check-free