The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves data from ServiceNow records and processes it to generate remediation plans, which creates an attack surface for indirect prompt injection.
Ingestion points: The skill queries tables such as sn_grc_issue, sn_grc_risk, and sn_compliance_control (SKILL.md, Step 1 and Step 2).
Boundary markers: No specific delimiters or instructions are provided to the agent to ignore or isolate potentially malicious instructions within the data retrieved from the records.
Capability inventory: The skill has access to high-privilege tools including SN-Execute-Background-Script, SN-Update-Record, and Bash (SKILL.md, Tools section).
Sanitization: No sanitization or validation logic is defined to check record fields (like short_description or description) before they are used in the generation of action plans or compliance tasks.
[DYNAMIC_EXECUTION]: The skill uses the SN-Execute-Background-Script tool to run server-side JavaScript (GlideRecord API) within the ServiceNow environment.
Multiple examples (Step 3, Step 4, and Example 2) demonstrate the assembly of scripts at runtime using templates where record identifiers or other data points are interpolated. While these templates are intended for data aggregation and bulk updates, the tool itself represents a high-capability execution vector if the agent is misled by malicious input.
[COMMAND_EXECUTION]: The skill utilizes Bash as a native tool and includes instructions for using curl to interact with ServiceNow REST API endpoints (SKILL.md, Step 1, Step 2, and Step 5).

issue-action-plan