Программа

Только названия модулей и уроков. Полный текст — после записи.

1. Module M01

   1. Lesson M01-L01 — What an agent is: model + tools + state + control loop
   2. Lesson M01-L02 — Agent vs deterministic workflow; observations and actions
   3. Lesson M01-L03 — Termination, budgets, bounded iteration
   4. Lesson M01-L04 — Traces, failure modes, when NOT to use an agent

2. Module M02

   1. Lesson M02-L01 — Task decomposition and plans as proposals
   2. Lesson M02-L02 — Replanning, state machines, working memory
   3. Lesson M02-L03 — Durable vs ephemeral state; context; checkpoints; resumability
   4. Lesson M02-L04 — Stop conditions and loop detection

3. Module M03

   1. Lesson M03-L01 — Reuse the JA-18 tool boundary inside agents
   2. Lesson M03-L02 — Permission scopes, allowlists, read/write, HITL for high impact
   3. Lesson M03-L03 — Prompt injection, tool-result injection, untrusted external content
   4. Lesson M03-L04 — Retries, idempotency, timeouts, unknown outcomes, safe recovery

4. Module M04

   1. Lesson M04-L01 — Success criteria, trajectory evaluation, tool-call accuracy
   2. Lesson M04-L02 — Safety evaluation, cost/latency, regression tests, fixtures
   3. Lesson M04-L03 — Production readiness for bounded agents
   4. Lesson M04-L04 — Ship JP-JA19 and transition toward JA-21 orchestration