Agents
Agents are autonomous instruction files that tell an AI assistant who it is, what its mission is, and how to behave. Drop them into a project and the agent runs until the work is done.
Development Agents
| Agent |
Purpose |
| AutoDev |
Senior developer — reads TODO.md and executes tasks autonomously |
| AutoDev E2E |
End-to-end testing — runs tests and validates the full stack |
| AutoDev Product Owner |
Product owner — analyses the codebase and fills TODO.md with the next batch of tasks |
Specialist Agents
| Agent |
Purpose |
| Architect |
System design, ADRs, C4 diagrams, scalability and failure mode analysis |
| Security Auditor |
OWASP Top 10, threat modelling, dependency CVE scanning, secrets detection |
| Database Engineer |
Schema review, index strategy, query optimisation, safe migration planning |
| QA Engineer |
Test strategy, risk-based coverage, test quality audit, edge case planning |
| DevOps Engineer |
CI/CD pipeline, containerisation, deployment automation, runbooks |
| Performance Engineer |
Profiling, benchmarking, bottleneck identification, load testing |
| Tech Writer |
README, API docs, CLI reference, changelog, contribution guides |
| Debugger |
Root cause analysis, systematic investigation, regression tests, post-mortems |
Engineering Specialist Agents
| Agent |
Purpose |
| Frontend Engineer |
React/Vue/CSS, component systems, Core Web Vitals, a11y-aware UI |
| Data Engineer |
ETL/ELT pipelines, warehouse design, data-quality validation |
| ML Engineer |
Model serving, feature stores, MLOps, LLM integration |
| Platform Engineer |
IDP, Kubernetes, Terraform, golden paths, developer experience |
| Accessibility Engineer |
WCAG compliance, screen-reader testing, axe audits |
| Mobile Engineer |
React Native / Flutter cross-platform, offline-first, app-store delivery |
| Incident Responder |
On-call SRE, incident command, blameless post-mortems |
| UX Engineer |
Design systems, interaction patterns, CSS tokens, design–dev handoff |
How Agents Work
Agents are markdown files used as system prompts or instruction contexts. Give the file to your AI assistant (Copilot, Claude, etc.) and it will adopt the identity and follow the mission described.
Typical full-product development flow:
- Architect — design the system before building it
- AutoDev Product Owner — populate
TODO.md with the first batch of tasks
- AutoDev — execute the tasks
- Security Auditor — audit before every release
- QA Engineer — verify quality gates
- Tech Writer — document what was built
- Repeat from step 2