# ReactorJet

A public working library for people building agent systems. ReactorJet publishes practical guides, draft component interfaces, field notes, and inspectable evidence. It does not currently offer a finished platform or SDK.

## Guides

- [/guides](/guides): tutorials, focused how-to guides, and explanations grouped by reader goal.
- [/guides/concept/what-makes-software-autonomic](/guides/concept/what-makes-software-autonomic) — What makes software autonomic (draft): 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.
- [/guides/how-to/design-an-evaluation-the-agent-cannot-edit](/guides/how-to/design-an-evaluation-the-agent-cannot-edit) — Design an evaluation the agent cannot edit (draft): Separate evaluation assets and authority from the agent so a proposal cannot redefine the test that decides whether it passes.
- [/guides/tutorial/run-an-agent-task-with-an-independent-check](/guides/tutorial/run-an-agent-task-with-an-independent-check) — Run an agent task with an independent check (draft): Run an agent in a restricted workspace, submit its result to a separate checker, and keep only work supported by recorded evidence.
- [/guides/concept/control-plane-versus-work-plane](/guides/concept/control-plane-versus-work-plane) — Control plane versus work plane (draft): The control plane holds intent, policy, and authority while the work plane performs bounded actions and reports what happened.
- [/guides/how-to/define-a-policy-gate](/guides/how-to/define-a-policy-gate) — Define a policy gate (draft): Turn a written constraint into a separate, testable decision that permits or blocks a consequential agent action.
- [/guides/tutorial/add-a-human-approval-boundary](/guides/tutorial/add-a-human-approval-boundary) — Add a human approval boundary (draft): Split preparation from execution so an agent can assemble consequential work while a person controls whether that exact action proceeds.
- [/guides/concept/evidence-before-confidence](/guides/concept/evidence-before-confidence) — Evidence before confidence (draft): 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.
- [/guides/how-to/structure-evidence-for-a-workflow](/guides/how-to/structure-evidence-for-a-workflow) — Structure evidence for a workflow (draft): Build an evidence bundle that connects each workflow claim to a reproducible observation, its source, and the decision that used it.
- [/guides/tutorial/trace-a-failed-agent-run](/guides/tutorial/trace-a-failed-agent-run) — Trace a failed agent run (draft): Reconstruct a failed run from durable records, locate the first broken assumption, and choose a recovery without erasing the evidence.
- [/guides/concept/ai-operations-ime](/guides/concept/ai-operations-ime) — AI operations IME (draft): 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.
- [/guides/how-to/choose-where-humans-approve](/guides/how-to/choose-where-humans-approve) — Choose where humans approve (draft): Place human approval at the smallest number of consequential transitions where judgment, accountability, or exception handling is still required.

## Components

- [/components](/components): reusable workflow building blocks with explicit inputs, outputs, and limits.
- [/components/intent-brief](/components/intent-brief) — Intent brief (draft): A versioned task contract that tells a software factory what outcome to pursue, what boundaries to respect, and how completion will be judged.
- [/components/policy-gate](/components/policy-gate) — Policy gate (draft): A separate decision point that allows, denies, or escalates a proposed action against versioned rules before the action takes effect.
- [/components/evidence-packet](/components/evidence-packet) — Evidence packet (draft): A bounded manifest that ties a proposed result to the checks, artifacts, provenance, and known gaps needed for review.
- [/components/independent-evaluator](/components/independent-evaluator) — Independent evaluator (draft): A separately controlled judge that compares a proposed result and its evidence with the original intent and a versioned rubric.
- [/components/human-approval-checkpoint](/components/human-approval-checkpoint) — Human approval checkpoint (draft): A scoped handoff that gives an accountable person the evidence, options, and authority needed to approve or stop a consequential action.

## Reference (start here)

- [/reference/autonomic-software-engineering](/reference/autonomic-software-engineering): the canonical concept.
- [/reference/closure-test](/reference/closure-test): the four-condition rubric.
- [/reference/two-plane-loop](/reference/two-plane-loop): the reference architecture.

## Packs (installable)

- [/packs/autonomic-swe/agent-eval](/packs/autonomic-swe/agent-eval) — agent-eval (0.2.0, hardened): Closes the loop on agent evaluation by separating the Verdict plane from the Proposal plane, so a change to the corpus can never silently tune the judge that decides whether the change helped.

## Reference index

- [/reference](/reference): A–Z plus domain grouping.

## Notes

- [/notes/control-loop/a-passing-test-is-not-a-good-result](/notes/control-loop/a-passing-test-is-not-a-good-result) — A passing test is not always a good result (control-loop, 2026-08-12): Tests tell an agent whether known checks passed; they do not automatically prove that the change helped users, respected every boundary, or is safe to keep.
- [/notes/control-loop/agent-activity-is-not-progress](/notes/control-loop/agent-activity-is-not-progress) — Agent activity is not progress (control-loop, 2026-08-13): Tool calls and busy run logs measure agent activity; progress begins when the workflow can prove that the target state moved closer and stayed there.
- [/notes/control-loop/approval-fatigue-is-a-workflow-design-problem](/notes/control-loop/approval-fatigue-is-a-workflow-design-problem) — Approval fatigue is a workflow design problem (control-loop, 2026-08-13): When an agent asks for approval too often and without decision-ready context, the workflow trains people to click through instead of protecting the actions that deserve judgment.
- [/notes/control-loop/closing-the-loop](/notes/control-loop/closing-the-loop) — Build the checks before you build the agent (control-loop, 2026-08-06): Decide how you will judge the result before building the agent, or your examples and tests will slowly start to favor whatever the agent already does well.
- [/notes/control-loop/rollback-belongs-in-the-plan](/notes/control-loop/rollback-belongs-in-the-plan) — Rollback belongs in the plan (control-loop, 2026-08-13): An agent should know how a consequential action will be contained or reversed before it acts, because a rollback invented after failure is only another untested production change.
- [/notes/control-loop/your-agent-is-creating-evidence-debt](/notes/control-loop/your-agent-is-creating-evidence-debt) — Your agent is creating evidence debt (control-loop, 2026-08-13): When an agent changes a system faster than you preserve why its work was accepted, each successful run leaves future operators with uncertainty they will have to repay.
- [/notes/control-loop/your-agent-should-not-be-its-own-reviewer](/notes/control-loop/your-agent-should-not-be-its-own-reviewer) — Your agent should not be its own reviewer (control-loop, 2026-08-11): Let an agent explain and improve its work, but use separate evidence and authority for the decision to accept, publish, merge, or deploy it.
- [/notes/teardown/scoring-cursor](/notes/teardown/scoring-cursor) — Can Cursor prove its own work? (teardown, 2026-08-06): Cursor can write code, run tests, and iterate, but a person still approves the work and the same agent that makes the change also interprets whether it worked.

## Work

- [/work/reconciler](/work/reconciler) — control-loop / reconciler (exploratory): Reference instantiation of the two-plane loop on a single repository. Repository repair per blueprint §6, blast radius L1 (Gated).

## Tools

- https://canvasui.dev — canvas-ui: The visual-direction reference for this lab site. We are not the maintainers; the lab holds the design tokens and the component shapes, and adapts them to Astro + Tailwind v4. We use the Particle Reveal effect on the home page hero and the Glitch effect on 404 / empty facets. The 80 KB effect budget on the home page is the binding constraint that decides which canvas-ui components we ship; the rest are out of scope.

## People

- [/people/ada](/people/ada) — Ada
- [/people/grace](/people/grace) — Grace
- [/people/tim](/people/tim) — Tim

## Directories

- [/guides](/guides): task-oriented documentation.
- [/components](/components): software factory building blocks.
- [/notes](/notes): opinions, lessons, and teardowns.
- [/work](/work): project directory with maturity badge.
- [/packs](/packs): pack directory with install command.
- [/tools](/tools): tools the lab has a position on.
- [/people](/people): team.
- [/participate](/participate): four shapes of contribution.