Build with ReactorJet
Start with what you need to do.
Learn through a small task, solve a specific workflow problem, or understand the model behind it. These drafts are public so you can use them while the library develops.
01 / Tutorials
Learn by completing one small workflow
Follow a guided path from setup to a result you can inspect.
- 01draft
Run an agent task with an independent check
Run an agent in a restricted workspace, submit its result to a separate checker, and keep only work supported by recorded evidence.
Outcome
A task run whose proposal and verdict come from separate authorities- agent-systems
- evals
- 02draft
Add a human approval boundary
Split preparation from execution so an agent can assemble consequential work while a person controls whether that exact action proceeds.
Outcome
A fail-closed approval step tied to one reviewed action and its evidence- agent-systems
- trust-boundary
- 03draft
Trace a failed agent run
Reconstruct a failed run from durable records, locate the first broken assumption, and choose a recovery without erasing the evidence.
Outcome
A causal failure report linked to the task, proposal, evidence, verdict, and cleanup state- agent-systems
- runtime-infra
02 / How-to guides
Solve one design problem
Use a focused procedure when you already know the outcome you need.
- 01draft
Design an evaluation the agent cannot edit
Separate evaluation assets and authority from the agent so a proposal cannot redefine the test that decides whether it passes.
Outcome
A protected evaluation with explicit ownership, fixed inputs, reproducible scoring, and a tested failure path.- agent-systems
- evals
- 02draft
Define a policy gate
Turn a written constraint into a separate, testable decision that permits or blocks a consequential agent action.
Outcome
A policy gate with bounded inputs, explicit decisions, separate authority, and tests for denial and failure.- agent-systems
- trust-boundary
- 03draft
Structure evidence for a workflow
Build an evidence bundle that connects each workflow claim to a reproducible observation, its source, and the decision that used it.
Outcome
A claim-linked evidence bundle that records provenance, versions, negative checks, and the final decision.- agent-systems
- evals
- 04draft
Choose where humans approve
Place human approval at the smallest number of consequential transitions where judgment, accountability, or exception handling is still required.
Outcome
An approval map that identifies which transitions need a person, what they review, and how approval is bound to the action.- agent-systems
- trust-boundary
03 / Concepts
Understand why the pattern works
Read the ideas behind control loops, evidence, boundaries, and agent operations.
- 01draft
What makes software autonomic
Software is autonomic when it can observe the result of its own action, judge that result against a protected target, and choose the next bounded step.
Outcome
Distinguish an autonomic control loop from automation, code generation, and unbounded autonomy- autonomic-swe
- agent-systems
- 02draft
Control plane versus work plane
The control plane holds intent, policy, and authority while the work plane performs bounded actions and reports what happened.
Outcome
Place decisions, permissions, execution, and evidence on the correct side of an operational boundary- agent-systems
- runtime-infra
- 03draft
Evidence before confidence
Confidence describes a belief about a result, while evidence records observations that another person or system can inspect and use to decide what happens next.
Outcome
Replace self-reported confidence with an evidence chain suited to the consequence of the decision- evals
- agent-systems
- 04draft
AI operations IME
AI operations IME is an emerging ReactorJet and Biro direction for turning operator intent into inspectable, bounded work without pretending that a prompt is authority or evidence.
Outcome
Understand the IME analogy, the proposed interaction boundary, and what has not been built or released- agent-systems
- trust-boundary