The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where untrusted data from compliance documents could potentially influence agent behavior.
Ingestion points: The skill instructions in SKILL.md direct the agent to read existing control lists, company policies, audit calendars, and auditor request lists from the local environment.
Boundary markers: There are no technical boundary markers or specific delimiters defined to distinguish untrusted file content from the skill's operational instructions.
Capability inventory: The agent has the capability to write multiple files (control-map.csv, evidence-plan.csv, mapping-notes.md) and communicate with an external network endpoint via an MCP server.
Sanitization: The skill lacks technical sanitization for ingested content, relying instead on natural language instructions for the agent to filter sensitive data before making external requests.
[DATA_EXFILTRATION]: The skill establishes a network connection to an external service for lookup operations.
Network Activity: The skill instructs the agent to use an MCP server at https://account.ismscopilot.com/v1/account/mcp to perform framework lookups.
Context: While the instructions explicitly state that the agent should never include file contents, secrets, personal data, or evidence in these requests and must show the question to the user first, the presence of a network path coupled with access to sensitive internal documents remains a monitored behavior.

multi-framework-overlap-map