ai-agent-evaluation-benchmarking – skill Agent & Tự động hóa cho agent AI

AI Agent Capability Evaluation & Benchmarking

What it is: Standardizes multi-metric capability benchmarking, token cost efficiency, and regression monitoring across autonomous coding agents.

Overview

Provides a standardized, auditable framework and data model for AI Agent Capability Evaluation & Benchmarking operations across distributed engineering and decentralized application systems.

When to Use This Skill

• When recording agent evaluation runs with a fixed dataset, rubric, runtime/model revision and measured outcomes.
• When comparing a candidate with a baseline under matched tasks, environment, tool permissions and cost accounting.
• When generating reproducible CSV, SQL DDL, JSON Schema, and Notion property registers for tracking compliance.

How It Works

1. Define the parameters, thresholds, and identity bindings required for the target operational register.

2. Select appropriate boundary enforcement values from validated enum select sets.

3. Export standardized artifacts (CSV table, SQL DDL, JSON Schema) to integrate into validation CI pipelines.

Field Reference

| # | Field Name | Type | SQL Type | JSON Schema Type | Notion Property Type | Example Value |

|---|------------|------|----------|------------------|----------------------|---------------|

| 1 | Benchmark Run ID | id | SERIAL PRIMARY KEY | integer | Text | 1 |

| 2 | Evaluated Agent Model | select | VARCHAR(64) | string | Select | Claude 3.7 Sonnet |

| 3 | Benchmark Suite Domain | select | VARCHAR(64) | string | Select | SWE-bench Verified |

| 4 | Tasks Evaluated Count | number | INTEGER | integer | Number | 100 |

| 5 | Pass Rate Percentage | number | NUMERIC(5,2) | number | Number | 78.40 |

| 6 | Tool Hallucination Rate % | number | NUMERIC(5,2) | number | Number | 0.60 |

| 7 | Average Tokens Per Task | number | INTEGER | integer | Number | 42500 |

| 8 | Cost Per Solved Task USD | currency | NUMERIC(8,4) | number | Number | 0.3420 |

| 9 | Regression Verdict | select | VARCHAR(32) | string | Select | Improved with evidence |

| 10 | Evaluation Lead | text | VARCHAR(64) | string | Text | Ranjeet2063 |

| 11 | Benchmark Execution Date | date | DATE | string, format: date | Date | 2026-10-01 |

Select Options

Evaluated Agent Model

Use the exact evaluated model/provider/deployment revision; examples are not an availability allowlist.

Benchmark Suite Domain

SWE-bench Verified | WebArena | AgentBench | HumanEval-Rust | Web3AuditBench

Regression Verdict

Improved with evidence | No demonstrated change | Regressed with evidence | Inconclusive

Measurement procedure

Record an immutable run identifier, suite revision, task IDs, model/runtime/tool settings, completed/failed/skipped counts and actual charges. Derive pass rate from eligible completed outcomes under a declared denominator; tool hallucination needs a defined invalid-call denominator. For a regression verdict, use paired outcomes or a justified test and interval, and retain an inconclusive state. Labels and example values below are synthetic, not measured results. Define any relation field in every emitted format only when the target exists and is confirmed.

Relations

• Audit Reference -> links to the formal review documentation or test repository.
• Target Architecture -> links to the deployed contract or autonomous agent runtime component.

Examples

Prompt

How do I configure and track AI Agent Capability Evaluation & Benchmarking for our production environment?

Recommended Next Step

> Generate the unified field schema, SQL DDL migration, and JSON validation schema to register into your system catalog.

>

> Workflow: Define criteria -> Run automated verification -> Record baseline -> Monitor invariants.

Best Practices

• Enforce strict typing on numerical bounds and currency amounts; avoid unstructured free-text fields for critical states.
• Re-run validation test suites on every state-altering commit or parameter change.
• Keep example data synthetic and isolated from production cryptographic keys or private endpoints.

Limitations

• This register does not execute benchmarks, prove compliance or need on-chain RPC credentials. An actual authorized harness and evaluation evidence are external prerequisites.
• Scores are comparable only with matched suite/revision, eligibility, failure/timeout handling and sampling. Record counts and uncertainty, not merely a percentage.
• Define cost per solved task as total attributable run cost divided by solved tasks; zero solves means unavailable/infinite cost under an explicit policy, not zero. Token estimates alone omit provider/tool charges.

Security & Safety Notes

• Generating a register is distinct from running paid providers, transmitting private tasks, changing CI or deploying; those effects need actual authorization.
• Use synthetic dummy keys and mock addresses in test suites and local verification scripts.

Common Pitfalls

• Problem: Mismatched decimal precision between contract runtime and database register.

Solution: Always verify decimals using the explicit field mapping in this reference.

• Problem: Missing authorization checks prior to state update.

Solution: Cross-validate against the Security Audit register before deployment.

Related Skills

• @ai-agent-tool-routing - covers tool schema registration and retry policy.
• @ai-prompt-regression-testing - covers prompt regression baselines and drift.
• @ai-code-generation-guardrails - covers static guardrails for generated code.

Reusable Prompt

I want to establish a verified AI Agent Capability Evaluation & Benchmarking register for our production protocol.
Guide me through the required field parameters and output the corresponding SQL DDL and JSON Schema.