AGT-01 · Agent templates: choose, build, run and upgrade¶
Teams start each agent from a blank file, so state, retry and termination are rebuilt differently every time and agents loop or stall in production. When something goes wrong nobody knows which template, framework version or prompt is running, so triage and upgrades take days.
- Who reads it: GenAI engineers building agents and DevOps engineers running them in production
- Outcome: Every agent starts from a tested template with state, retry and stop rules
- Pack owner: James (catalog owner role: Agent Lead)
Master catalog entry¶
- Catalog name: Agent Template Library (P1)
- Why it matters: Agentic work commands higher rates and longer engagements
- What it removes: Task, planner, supervisor and tool agents ready to specialise
- Stack: LangGraph, Semantic Kernel, Azure AI Foundry Agent Service
- Maturity target: Level 3 — six templates with state, retry and termination built in
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-AGT-01-01 | AGT-01-S1 | Choose the agent template and runtime (LangGraph, Agent Framework or Foundry Agent Service) | Decision guide | Medium | 3 | draft | James | 2026-11-26 |
| PB-AGT-01-02 | AGT-01-S2 | Build a task agent from the template with state, retry and termination | How-to | Medium | 3 | draft | Sanjay Sarvesh | 2026-12-08 |
| PB-AGT-01-04 | AGT-01-S3 | Upgrade the agent framework or model version without regressions | Runbook | High | 4 | draft | James | 2026-11-20 |
| PB-AGT-01-05 | AGT-01-S4 | Agent go-live checklist | Checklist | Low | 4 | draft | Sanjay Sarvesh | 2026-11-18 |