Skip to content

AI SDLC: Context-as-Code

A framework for steering AI agents (Cursor, Claude, Antigravity) to produce consistent, high-quality code by treating context as code.

Core Concept

Instead of repeating instructions in every chat session, we commit markdown configuration files to the repository. These files define the rules, skills, and current state for the AI.

1. Rules

Rules provide system-level instructions to the Agent. They can be scoped to specific projects or applied globally.

Types: - Project Rules (.cursor/rules/*.md): Version-controlled, scoped by path patterns (globs). Best for project-specific conventions. - User Rules (~/.cursor/rules / Dashboard): Global to your environment. Best for personal preferences. - Team Rules (Dashboard): Enforced for all team members.

Best Practices: - Keep rules focused and under 500 lines. - Reference files instead of duplicating code. - Avoid vague instructions; be concrete.

2. Skills (Tool Use)

Skills extend the agent's capabilities by providing it with "tools" (functions) it can call to interact with external systems or perform complex logic.

Concept:

  • Tool Definitions: JSON schemas defining the function signature (name, description, parameters).
  • Logic: The actual code (e.g., Python script, API call) that executes when the tool is called.

Claude/MCP Integration:

  • Define tools using the Model Context Protocol (MCP) or Claude's native tool format.
  • Agents can "call" these tools to fetch documentation, query databases, or run commands.
  • Example: A "Migrate to Helm" skill might be a composite of file operations and helm lint commands exposed as a comprehensive tool or workflow.

3. Context: PRDs & RFCs

Instead of maintaining a scratchpad, provide context through structured documentation.

Usage

  • PRDs (Product Requirement Docs): Pass the full PRD to the agent at the start of a feature.
  • RFCs (Request for Comments): Use technical design docs to align the agent on architectural decisions.

Workflow

  1. Prepare: Ensure your PRD or RFC is comprehensive and up-to-date.
  2. Inject: @mention the relevant PRD/RFC file in the chat or include it in the context window.
  3. Instruct: Ask the agent to "Implement the feature described in @prd.md" or "Review this code against @rfc.md".

4. Commands (/command)

Custom commands allow you to trigger reusable workflows with a simple / prefix in the chat.

Locations: - Project: .cursor/commands (Version controlled). - Global: ~/.cursor/commands (Personal). - Team: Managed in Cursor Dashboard (Shared).

Usage: - /doc: "Read the code and generate Javadoc/PyDoc." - /test: "Generate unit tests for the selected code." - With Parameters: /commit "fix login bug" (Passes context to the command).

5. MCP Support

Model Context Protocol (MCP) provides a standard interface for agents to connect with external data and tools.

Category Supported Integrations
Documentation Google Docs, Google Drive (Read specs, access loose notes).
Code & Deploy GitHub (PRs, Issues, Actions), Docker (Container management).
Project Management JIRA, Linear (Ticket tracking, sprint planning).
Communication Slack (Team updates, incident alerts).

MCP Setup Config

Add these to your MCP configuration file (e.g., claude_desktop_config.json).

Google Docs & Drive: 

"google-drive": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-google-drive"]
}

GitHub: 

"github": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": {
    "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
  }
}

Docker: 

"docker": {
  "command": "docker",
  "args": ["run", "-i", "--rm", "mcp/docker"]
}

JIRA: 

"jira": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-jira"],
  "env": {
    "JIRA_API_TOKEN": "<YOUR_TOKEN>",
    "JIRA_EMAIL": "<YOUR_EMAIL>",
    "JIRA_INSTANCE_URL": "<YOUR_INSTANCE_URL>"
  }
}

Linear: 

"linear": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-linear"],
  "env": {
    "LINEAR_API_KEY": "<YOUR_KEY>"
  }
}

Slack: 

"slack": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-slack"],
  "env": {
    "SLACK_BOT_TOKEN": "<YOUR_TOKEN>",
    "SLACK_SIGNING_SECRET": "<YOUR_SECRET>"
  }
}

Workflow

  1. Start: Agent reads .cursorrules (Style) and consumes context from PRDs/RFCs.
  2. Execute: Agent uses Skills and MCP Tools to perform tasks.
  3. Commit: Agent validates changes against the PRD/RFC requirements.