Documentation index

BoundaryKit is a documentation-first, vendor-neutral framework informed by a running local implementation of untrusted, tool-wielding agents. The case study describes that operational experience; the fictional worked example teaches the method without publishing private configuration. OpenShell and Ansible are used locally, but are not prerequisites for adopters.

Main reading path

1. Adoption guide — five stages, each with inputs, steps, a named output, and deterministic design/runtime gates.
2. Worked fictional example — a filled teaching packet; NOT RUN applies to its example probes, not to the running local implementation.
3. Architecture overview — trust boundaries, logical grouping versus enforced isolation, and the operational implementation example.
4. Governed workload case study — experience from the running local setup, its controls, and evidence scope.

Apply and review the claims

• Verification model — distinguish narrative, static checks, fictional examples, historical scoped observations, and production evidence.
• Threat model and security methodology — assets, attacker paths, and control assumptions.
• Governance and claim discipline — tiers, canaries, tool and provider objectives, and promotion gates.
• Promotion control plane — dry-run/review and rollback design.
• Gated preview access — temporary access pattern; separate authorization required.
• Public reader quickstart — safe static documentation checks, not runtime acceptance.

Historical evidence and design appendices

These pages support particular earlier claims or preserve retired acceptance-design lessons. They are neither a required stack nor maintained runnable fixtures; transfer no result to a new workload without a matching test.

• Boundary receipt #1 — historical scoped inner-sandbox refusals, not current runtime or grouping isolation.
• Boundary receipt #2 — historical placeholder configuration and one stopped misroute, not successful credential mediation.
• Historical isolation substrate — golden VM, signed-image, and microVM acceptance concepts.
• Historical gateway runtime layout — legacy per-profile gateway design, not a mapping to the operator's logical groupings.
• Illustrative acceptance receipt — fictional PASS examples, not measurements.

Public/private boundary

• Evidence model
• Public/private boundary
• Declassification checklist

The operator confirms that the local OpenShell implementation is running, with Ansible configuration as code supporting its workflows. The public account, recorded 2026-10-03, describes that experience without disclosing private topology or operational identifiers. Operation, a specific boundary test, and a measured recovery drill are different facts; each test claim stays tied to its own evidence. The fictional walkthrough does not describe the status of the real implementation. No managed service, client validation, or production certification is claimed.