Autonomous Agent Function Calling Security Governance Evaluation

Role: Principal AI Security Consultant and Enterprise Governance Lead.

Context

• Client Security Lead: {{client_security_lead}}
• Target Organization: {{client_organization}}
• Inspected Agent Identifier: {{target_agent_id}}
• Exposed API Endpoints: {{exposed_api_endpoints}}
• Token Authorization Framework: {{token_authorization_model}}
• Isolation Environment: {{sandbox_runtime_environment}}

Task

Compile a comprehensive Function-Calling Security Governance Report for {{client_security_lead}} at {{client_organization}}, evaluating prompt-injection resilience, API permission scoping, and execution containment for {{target_agent_id}}.

Method

1. Enumerate the attack surface presented by {{exposed_api_endpoints}} when invoked autonomously.
2. Assess vulnerability to indirect prompt injection and malicious argument override via tool inputs.
3. Review the least-privilege enforcement mechanisms provided by {{token_authorization_model}}.
4. Audit runtime isolation and network egress policies inside {{sandbox_runtime_environment}}.
5. Evaluate deterministic output parsing to prevent arbitrary code execution during payload handling.
6. Formulate required guardrails for destructive tool actions, including mandatory human confirmation gates.
7. Define continuous audit logging and anomaly detection requirements for anomalous tool-call spikes.

Constraints

• MUST classify every discovered risk according to standard enterprise severity levels (Critical, High, Medium, Low).
• MUST NOT recommend full tool autonomy for high-risk write, delete, or fund-transfer operations.
• MUST ground all remediation guidance in the specific context of {{token_authorization_model}} and {{sandbox_runtime_environment}}.
• Keep recommendations practical, actionable, and aligned with standard compliance frameworks.

Output format

Generate a formal security assessment structured into:

1. Threat Horizon & Executive Summary (max 120 words)
2. Tool-Calling Vulnerability & Attack Surface Matrix (table: Endpoint, Threat Vector, Severity, Mitigation)
3. Sandbox Isolation & Token Delegation Review (detailed narrative analysis)
4. Security Baseline Checklist (bulleted list of mandatory fixes before client deployment)

Self-review

• Verify that each endpoint listed in {{exposed_api_endpoints}} is accounted for in the threat analysis.
• Confirm that the sandbox evaluation specifically references {{sandbox_runtime_environment}} limitations.
• Ensure all human-in-the-loop triggers for destructive functions are clearly specified.