Building Reliable AI Coding Agents with Skills, MCP and Guardrails Training Course

From Autocomplete to Agents: Understanding Why Agents Fail

•     Anatomy of a coding agent: model, harness, tool interface, context, and permissions

•     Positioning of each tool: Claude Code, GitHub Copilot, Cursor, Codex CLI, Gemini CLI

•     A classification of failure modes: incorrect context, mismatched tools, lack of feedback, and unbounded autonomy

Demonstration: Comparing the same task executed effectively versus poorly

Context Engineering

•     Treating the context window as a resource budget: prioritizing valuable content

•     AGENTS.md, CLAUDE.md, .cursor/rules, copilot-instructions.md — a unified concept under different filenames, serving as a single source of truth

•     Defining conventions, build and test commands, and architectural boundaries

•     Retrieval vs. explicit context; task decomposition and sub-agent strategies

Lab: Author repository context for an unfamiliar Python service, then re-execute a previously failing task to compare results

Reusable Workflows and Agent Skills

•     Selecting the right abstraction level: instruction files, skills, custom commands, or standard scripts

•     Structure of a skill: triggering mechanisms, instructions, bundled scripts, and progressive disclosure

•     Cross-tool portability and identifying points of vendor lock-in

•     Version control, review processes, and team distribution; identifying common anti-patterns

Lab: Create and test a reusable workflow that enforces internal coding standards

MCP: Integrating Agents with Real-World Systems

•     Architecture overview: clients, servers, tools, resources, and prompts; utilizing stdio and HTTP transports

•     Justifying specific servers: Git hosting, issue tracking, databases, browsers, and internal APIs

•     Scenarios where a CLI or script outperforms an MCP server

•     Maintaining tool surface hygiene: why an excess of tools reduces reliability

Lab: Connect MCP servers to manage a ticket end-to-end — from issue creation to branch, patch, testing, and pull request

Feedback Loops and Evaluation

•     Using tests, types, and linters as the agent’s ground truth; employing test-driven approaches as a control mechanism

•     Leveraging CI as the external feedback loop and maintaining review discipline for agent-generated diffs

•     Creating golden-task evaluation sets: defining metrics and detecting regressions

•     Monitoring cost and latency as primary performance indicators

Lab: Construct a small evaluation set and benchmark two different agent configurations against it

Security and Guardrails

•     Mitigating prompt injection risks from issues, pull requests, READMEs, dependencies, and retrieved content

•     Implementing permission models: allowlists, approval workflows, read-only tools, and network egress controls

•     Maintaining secret hygiene and sandboxing: using containers, ephemeral credentials, and limiting potential impact

•     Assessing supply chain risks associated with third-party MCP servers and shared skills

Lab: Observe an agent being compromised by a malicious repository, then harden the configuration to prevent recurrence

Team Adoption and Rollout

•     Developing a phased adoption strategy; determining what to standardize and what to leave to individual discretion

•     Identifying metrics that reflect true value versus those that do not