Skip to content

Certify the adversarial implementation workflow end to end #299

Description

@taras

Story

As a release owner, I want an adversarial cross-runtime certification matrix, so completion of #181 is demonstrated under interruption, replay, stale state, and failure rather than inferred from happy paths.

Acceptance matrix

  • Process exit and partial replay against the same retained Workspace root.
  • Completed replay with zero Workspace, Agent, Worker, or external-provider attachment.
  • Restart while suspended for exact correlated user input.
  • Cancellation during Agent, filesystem, local Git, Worker Shell, forge, and user-input operations.
  • Atomic publication or rollback of local mutations and filtered journal results.
  • Missing/corrupt authoritative Workspace failure without false restoration.
  • Multi-repository and named-Worktree retention, including dirty and unpushed state.
  • Stale base, moved remote branch, changed pull-request head, and ambiguous external completion.
  • Mandatory read-only Agent authority and constrained generated-XMD preflight.
  • Malformed Agent output and exhausted correction or implementation attempts.
  • Duplicate-effect prevention, history inspection, compatible fork, and incompatible-fork rejection.
  • Secret rejection before journal persistence without scanning arbitrary Workspace contents.
  • Deno, Node, Bun, compiled binary, TypeScript, and JSR gates.
  • No required state exists only in an Agent transcript, host path, provider handle, or Git sidecar ref.
  • The final documents use only architecture.md terminology and describe shipped behavior.

Exit

Every required issue in #218's implementation map is closed, every end-goal capability in #181 is complete, and the living workflow matches production behavior. xmd play, a public remote-host selector, Worker JavaScript, native subprocesses, and writable FUSE remain separate later layers.

Dependencies

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions