Skip to content

Phase 3: make the admission demo self-verifying #54

Description

@johnnygreco

Objective

Make the real example self-verifying and bring user-facing/architecture documentation and bridge encoding in line with the implemented admission model.

Context

Scope

  • Demo verify cases: denied prompt/history invariant, redacted prompt, raw provider request without handle, stock Pi without runtime extension, and best-effort tool cases.
  • Add content-safe JSON logging sufficient to assert negative-control reason codes.
  • Stop exposing readiness metadata to workload.
  • Coordinate bridge JSON base64 request/replacement bodies across Research runtime adapter and OpenShell bridge.
  • Add Egress Gate admission architecture page and update README/current-scope/limits/denial/lifetime/dedicated-instance docs.
  • Correct OpenShell middleware and sandbox architecture docs.
  • Non-goals: caller tokens (Phase 4), mock runners, production deployment automation.
  • Dependencies: Phase 2.

Acceptance Criteria

  • ./demo.sh verify exercises real sandbox/Pi behavior and prints one result per case. Implementation is complete; execution is externally blocked by the absent prepared ignored runtime/CLI, .env, JSON log, and provider credential, as recorded in the completion comment and PR Review Record.
  • Number-array bridge bodies are gone.
  • Docs describe the actual three checkpoints, limits, reason codes, and residual risks.
  • Generated documentation mirror is produced through the staging workflow.

Required Validation

Research make check, docs renderer/build, focused OpenShell bridge tests/docs checks, and a real clean demo run when environment/provider access is available.

PR Expectations

Include sanitized verify output or explicitly record the external credential/runtime blocker. No mock fallback.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions