The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes campaign performance data from external sources (MCP connections or user-provided files). This data is passed to utility scripts for analysis, which creates a surface for indirect prompt injection if campaigns, ad groups, or other data fields contain malicious instructions intended to influence the agent's behavior during the audit.
Ingestion points: Data fetched via mcp__*__natural_to_data and manual file reads or pastes in SKILL.md (Path B).
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill has access to the Bash tool to execute Python scripts, which is used for data validation and metric calculation.
Sanitization: The instructions do not specify any sanitization or validation logic for the content of the data fields before processing.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the !cat command within the markdown body to load business context from a local file (.agents/product-marketing-context.md) at runtime. This is a standard use of project-specific tooling for context loading and does not involve user-controlled input or exfiltration.
[COMMAND_EXECUTION]: The skill executes a specific Python utility script (ds_utils.py) via the Bash tool to perform data processing tasks such as trend comparison and budget efficiency checks. The scope of execution is restricted by the allowed-tools frontmatter field to the specific script provided by the vendor.

ds-paid-audit