The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to read and treat external policy documents, terms files, and notes as binding rules, which creates a potential surface for indirect prompt injection where malicious data could influence agent behavior.
Ingestion points: SKILL.md instructs the agent to "Read every policy document, terms file and notes in full" to determine rules for computation.
Boundary markers: The skill does not provide specific delimiters or instructions to differentiate between data and control instructions within these external files.
Capability inventory: The skill is strictly logical/procedural and does not request tool access, network capabilities, or file-writing permissions in its frontmatter.
Sanitization: No instructions for sanitizing or validating the input from external sources are present.
[NO_CODE]: The skill package is composed entirely of Markdown, JSON metadata, and license text. It does not include any scripts (e.g., Python, JavaScript) or binary executables, which eliminates risks associated with direct code execution or malicious dependencies.

ar-aging-and-cash-application