Agent instructions
AuraScore 83/100

Enterprise API Agent Tool-Use Protocol Audit

Evaluate autonomous tool-selection instructions, API calling contracts, and error recovery policies for production agent systems.

Use this template when preparing autonomous developer platform agents for production deployment or auditing tool-invocation reliability. It systematically reveals edge-case handling bugs, hallucinated parameters, and auth-boundary failures.

Template

Role: Principal AI Architect and Systems Integration Lead with fifteen years of experience designing distributed systems and autonomous tool-execution engines.

Context

  • Target deployment platform: {{platform_name}}
  • Integrated API ecosystem: {{target_api_suite}}
  • Active system instructions: {{current_system_prompt}}
  • Agent memory and state architecture: {{state_management_model}}
  • Identified high-risk failure modes: {{critical_failure_scenarios}}
  • Authorization boundaries: {{auth_scope_boundaries}}

Task

Deliver an exhaustive architectural analysis of the agent's tool-calling instructions to eliminate non-deterministic parameter hallucination, enforce strict schema adherence, and guarantee graceful fallback behaviors during upstream API failures.

Method

  1. Deconstruct {{current_system_prompt}} to extract implicit assumptions regarding API payload construction and output parsing.
  2. Map every tool declared in {{target_api_suite}} against the validation logic defined in the agent instructions.
  3. Evaluate the agent's state-tracking instructions under {{state_management_model}} for token drift, context starvation, and duplicate action loops.
  4. Stress-test the instructions against {{critical_failure_scenarios}}, specifically analyzing how the agent handles 4xx/5xx network errors and schema drift.
  5. Audit instruction-level authorization enforcement to prevent privilege escalation beyond {{auth_scope_boundaries}}.
  6. Identify ambiguities where the agent might hallucinate optional arguments or execute destructive operations without confirmation.
  7. Formulate remediated instruction clauses with exact syntactic and structural constraints for deterministic execution.

Constraints

  • MUST evaluate tool-invocation schemas against zero-trust architectural principles.
  • MUST NOT suggest external human approval gates for operations explicitly within {{auth_scope_boundaries}}.
  • Analysis MUST explicitly differentiate between prompt-level vulnerabilities and API-gateway failures.
  • Recommendations must preserve token budget efficiency while maximizing deterministic behavior.

Output format

  1. Executive Protocol Vulnerability Scorecard (0-100 rating with critical risk callouts)
  2. Tool Schema & Parameter Rigor Analysis (bulleted breakdown per tool in {{target_api_suite}})
  3. State & Loop Prevention Evaluation (technical assessment under {{state_management_model}})
  4. Failure Mode Simulation Matrix (table: Scenario, Failure Risk, Instruction Weakness, Mitigating Directive)
  5. Hardened Instruction Replacement Modules (production-ready prompt diffs in code blocks)

Self-review

  • Did I verify that all remediated prompt clauses address every edge case in {{critical_failure_scenarios}}?
  • Are all prompt modifications strictly compatible with {{platform_name}} formatting standards?
  • Have I ensured no sensitive authorization scopes in {{auth_scope_boundaries}} can be bypassed via prompt injection?
AuraScore breakdown
83/100Provisional
Instruction clarity15/15 · Strong

Explicit role, a named task, and discrete steps the model can follow.

Context architecture12/12 · Strong

Background, inputs and variables the model needs before it starts.

Constraint engineering12/12 · Strong

Hard boundaries — what the model must and must not do.

Output specification6/14 · Thin

A named, field-level shape for the response.

Reasoning structure10/10 · Strong

Ordered work items that force analysis before an answer.

Model compatibility10/10 · Strong

Length and structure that travel across frontier models.

Token efficiency5/10 · Thin

Signal density — instruction weight without padding.

Reusability7/7 · Strong

Documented variables so the scaffold adapts to new inputs.

Robustness5/5 · Strong

Quality bar, assumptions and behaviour when inputs are thin.

Observed performance1/5 · Thin

How much real usage the template has behind it.

ai-agents
agents-instructions
technology-software
tool-calling
api-agents
agent-instructions