Coding Agent

Give your agents full-stack software development capabilities — reading, writing, editing files, searching code, and executing commands — all within a sandboxed project directory.

Overview

The coding agent feature provides six MCP tools that let agents work with code the way a developer would. Agents follow a structured methodology: gather context, take action, verify results.

Agent receives coding task
    │
    ├── 1. Gather Context
    │   ├── coding_list_files   → Understand project structure
    │   ├── coding_search       → Find relevant files
    │   └── coding_file_read    → Read existing code
    │
    ├── 2. Take Action
    │   ├── coding_file_edit    → Modify existing files
    │   ├── coding_file_write   → Create new files
    │   └── coding_bash         → Run commands (with approval)
    │
    └── 3. Verify Results
        └── coding_bash         → Run tests, check build

Getting Started

1. Assign the Skill

The coding tools require the coding-agent skill. Assign it to your agent:

await client.skills.assign({
  skillId: 'coding-agent',
  agentId: 'agent-123'
});

Or let the platform suggest it automatically — connecting a GitHub integration triggers a suggestion for the coding-agent skill.

2. Agent Gets Coding Tools

Once the skill is assigned, the agent can use all six coding tools. The skill also provides methodology guidance that the agent follows when working on tasks.

Available Tools

coding_file_read

Read file contents with line numbers.

// Read entire file
{ name: "coding_file_read", arguments: { path: "src/index.ts" } }

// Read specific lines
{ name: "coding_file_read", arguments: { path: "src/index.ts", startLine: 10, endLine: 50 } }

Returns numbered content:

     1  import express from 'express'
     2  const app = express()
     3  ...

coding_file_write

Create a new file or overwrite an existing one.

{
  name: "coding_file_write",
  arguments: {
    path: "src/utils/helper.ts",
    content: "export function add(a: number, b: number) {\n  return a + b\n}\n"
  }
}

Parent directories are created automatically. Prefer coding_file_edit for modifying existing files.

coding_file_edit

Edit a file by replacing an exact string match.

{
  name: "coding_file_edit",
  arguments: {
    path: "src/index.ts",
    oldString: "const port = 3000",
    newString: "const port = process.env.PORT || 3000"
  }
}

The oldString must match exactly, including whitespace. If it appears multiple times, set replaceAll: true or provide more surrounding context to make it unique.

coding_search

Search file contents or find files by name.

// Search file contents (grep-like)
{
  name: "coding_search",
  arguments: { pattern: "handleAuth", type: "content", glob: "*.ts" }
}

// Find files by name (glob-like)
{
  name: "coding_search",
  arguments: { pattern: "*.test.ts", type: "files" }
}

coding_bash

Execute a shell command in the project directory.

{
  name: "coding_bash",
  arguments: { command: "npm test", timeoutMs: 60000 }
}

important

Every bash execution requires human-in-the-loop approval. The agent proposes a command, and you approve or reject it before it runs.

Common uses:

• Running tests: npm test, pytest
• Building: npm run build, cargo build
• Git operations: git diff, git status
• Package management: npm install, pip install

coding_list_files

List directory contents to understand project structure.

// List project root
{ name: "coding_list_files" }

// List a subdirectory recursively
{ name: "coding_list_files", arguments: { path: "src", recursive: true } }

Security & Sandboxing

All coding operations run inside a sandbox that enforces safety boundaries.

Project Directory Isolation

Every file operation is confined to the project directory. Path traversal attempts (../../etc/passwd) are blocked. The agent cannot read or write files outside its workspace.

Blocked Paths

Sensitive paths are blocked by default:

• System directories (/etc, /root, /usr, /var)
• Environment files (.env, .env.local, .env.production)
• Dependencies (node_modules)

Bash Safety

Destructive commands are blocked:

Blocked Pattern	Why
rm -rf /	Filesystem destruction
chmod 777	Insecure permissions
curl | sh	Remote code execution
Fork bombs	Resource exhaustion
shutdown, reboot	System control

Network access from bash is disabled by default. When enabled, only whitelisted hosts are accessible.

Audit Trail

Every operation is logged with timestamp, tool name, action, path/command, and whether it was allowed or blocked. Retrieve the full audit log at any time for compliance review.

File Size Limits

File read and write operations are limited to 1MB by default to prevent resource exhaustion.

Works with Any Model

The coding tools work with every LLM provider supported by MeetLoyd:

• Anthropic — Claude Opus, Sonnet, Haiku
• OpenAI — GPT-4o, o1, o3
• Google — Gemini 2.0
• Self-hosted — Ollama, vLLM (DeepSeek, Llama, Qwen, Mixtral)

The tools plug into the existing agent execution loop — no special model or SDK required.

Configuration

The sandbox can be configured per agent or per tenant:

Option	Default	Description
projectDir	/workspace	Root directory for all file operations
allowBash	true	Enable bash tool
allowNetwork	false	Allow network access in bash
allowedHosts	[]	Network whitelist when enabled
maxFileSizeKb	1024	Max file size for read/write
blockedPaths	System dirs, .env, node_modules	Additional paths to block
blockedCommands	Destructive patterns	Additional commands to block
auditEnabled	true	Full audit logging

Best Practices

Read Before Editing

Always read a file before modifying it. This ensures the agent understands context and gets the exact string to replace.

Small, Focused Changes

Make one logical change at a time. Large, multi-file changes are harder to verify and more likely to introduce errors.

Verify After Every Change

Run tests and check the build after each modification. Use coding_bash with npm test or your project's test command.

Match Existing Patterns

The agent follows the codebase's existing conventions for naming, structure, and style — don't fight the codebase.

Use Git

Leverage git for safety. git diff shows what changed, git stash provides a safety net, and git status gives a clear picture of the current state.

---

Next: Learn about Human-in-the-Loop controls for governing agent actions.