Skip to content

Capability Map

A competency map is not another set of courses, but a specialized view of the same content. It regards AI Coding as a complete engineering system to help platform builders and technical leaders identify the problems, boundaries and governance methods of each link.

Two-tier professional model 

Life cycle layer: input and context → Agent execution and capability expansion → output and verification
Governance layer: Model management · Gateway management · Cost analysis · Quality assessment
Across the entire life cycle

The first three items describe how an AI Coding task flows; the last four items are not subsequent steps, but governance dimensions that act on the entire process of input, execution, and output.

1. Input and context

Pay attention to whether the AI ​​obtains correct, sufficient and clearly bounded information.

  • Project rules: from Rules to AGENTS.md
  • Requirements specification: from natural language requirements to specification-driven development
  • Context Engineering: Context selection, compression and updating of large code bases

2. Execution and capability expansion

Pay attention to how Agents invoke tools, connect systems, reuse experiences, and complete complex tasks.

  • Connect MCP, Tools and external systems
  • From Prompt to Agent Skill
  • From Rules, Skills to Plugin
  • Complex task dismantling and multi-agent workflow

3. Output and verification

Pay attention to whether the generated results are correct, maintainable, and verifiable.

  • From code generation to verification engineering
  • Testing, static inspection, review and delivery evidence
  • Modification scope, regression risk and quality gate control

Horizontal governance layer

4. Model management

Pay attention to model selection, routing, capability boundaries and real engineering performance.

  • Model capability boundaries and task adaptation
  • Benchmark and real project evaluation
  • Model version changes and regression verification

5. Gateway management

Pay attention to the unified entrance, permissions, current limiting, routing and auditing of model calls.

  • Responsibilities and boundaries of AI Gateway
  • Multi-model routing, degradation and disaster recovery
  • Permissions, quotas and call auditing

6. Cost analysis

Focus on the true costs of tokens, tool calls, retries, and contextual strategies.

  • Consumption analysis of Claude Code and Codex
  • Task level cost attribution
  • Trade-off between quality, speed and cost

7. Quality Assessment

Watch how teams continue to measure whether AI coding is truly effective.

  • Code correctness and maintainability
  • End-to-end task success rate
  • Efficiency improvement, rework rate and manual intervention level -Team-level evaluation mechanism

How to coordinate the two routes

It is recommended to follow the Growth Route when learning for the first time; when designing a platform, troubleshooting system problems, or building team specifications, you can view a certain professional concern horizontally through this page.