Eslam HelmyEslam Helmy
•4 min read•Eslam

Build Your Own Dev Agent — Lesson 1: What Is a Dev Agent

claude-codedev-agentautomation

You are about to build a personal developer agent. Not a chatbot. Not an autocomplete tool. An autonomous system that runs on your machine, monitors your work, makes decisions, and reports back to your phone.

The Finished Agent

Here is the .claude/ directory of a fully operational dev agent:

.claude/
  CLAUDE.md                  # Master instruction file -- the agent's brain
  preferences.md             # Who you are, your calendars, your don'ts
  tasks-active.md            # Work in progress
  tasks-completed.md         # Done items
  progress.txt               # Append-only action log
  learnings.md               # Patterns, mistakes, preferences
  error-log.md               # Past mistakes -- never repeat
  priority-map.md            # 4 priority levels with rules
  cron-jobs.json             # 7 scheduled jobs
  failed-jobs.log            # Dead-letter queue
  hooks/
    stop-telegram.sh         # Sends notification when agent finishes
  skills/
    daily-planner/SKILL.md
    pr-reviewer/SKILL.md
    git-reviewer/SKILL.md
    standup-generator/SKILL.md
    meeting-ingest/SKILL.md
    learning-loop/SKILL.md
    browser-verify/SKILL.md
    heartbeat/SKILL.md

This agent has 7 scheduled jobs running. It sends Telegram notifications when tasks finish. It monitors GitHub repos, reviews PRs, generates standups, ingests meeting transcripts, and self-heals when something breaks. It learns from corrections and gets better every week.

The Autonomy Loop

Every agent action follows the same cycle:

Trigger --> Read State --> Decide --> Act --> Verify --> Update State --> Report
Autonomy Loop: Trigger → Decide → Act → Verify → Report
  1. Trigger -- A cron fires.
  2. Read State -- The agent reads its state files to understand the current situation.
  3. Act -- It executes: generates a draft, queries an API, writes code, runs a tool.
  4. Verify -- It checks its own work. Did the file get created? Does the output make sense?
  5. Update State -- It writes results to the appropriate state files. progress.txt gets a new line. tasks-active.md gets updated.
  6. Report -- It sends a notification. Telegram, Slack, or just a log entry.

Every skill you build in this course follows this loop.

What Makes This Possible

Three things make Claude Code suitable for this:

  • CLAUDE.md -- The master instruction file Claude Code reads at startup. It defines what to read, what rules to follow, and how to behave.
  • Hooks -- Shell scripts that fire on specific events, making behavior deterministic: notifications on completion, blocking dangerous operations.
  • File-based state -- All agent state lives in plain text files inside .claude/, readable, version-controlled, and portable.

The 7 Scheduled Jobs

Here is what the finished agent runs on autopilot:

  • Daily Planner (5:33 PM) -- Reviews calendar, scores the day, plans tomorrow
  • Git Reviewer (Noon) -- Summarizes commits across your repos
  • PR Reviewer (3x daily) -- Monitors open PRs, flags risks
  • Standup Generator (8 AM) -- Generates daily standup from tasks and progress
  • Meeting Ingest (6:37 PM) -- Extracts action items from meeting transcripts
  • Learning Loop (11:47 PM) -- Consolidates corrections into permanent rules
  • Heartbeat (Every 2h) -- Self-checks: crons alive, state valid, no stale tasks

Browser Verify is an on-demand skill -- called by other skills after changes, not on a schedule.

What You Will Build

Over the next 7 lessons, you will construct this entire system from scratch:

  • Lesson 2 -- CLAUDE.md and state files. The agent's foundation.
  • Lesson 3 -- Hooks. Telegram notifications and safety gates.
  • Lesson 4 -- Memory. Inline learning, error tracking, autonomy rules.
  • Lesson 5 -- Skills and scheduling. Your first cron job.
  • Lesson 6 -- Building skills. PR reviewer, git reviewer, standup generator.
  • Lesson 7 -- Meeting ingest and failure handling. Graceful degradation.
  • Lesson 8 -- Heartbeat, remote access, and ship it. The self-healing safety net, persistent sessions, and deployment.

By lesson 8, you will have built all of this.

This is part of the Build Your Own Dev Agent course. Next Lesson →

Share this post