0.1.8 Beta7: Add the versioned CLI-only release-verification profile #255

Parent

#319 (part of #248)

Outcome

Extend the existing real-client runner with one declarative, versioned 0.1.8 release-verification profile. Preserve the accepted 0.1.7 profile unchanged and do not introduce a second product acceptance harness.

Product seam

The real Codex CLI through the exact candidate Gateway is the 0.1.8 product-level acceptance seam. Synthetic/protocol-controlled fixtures verify generic behavior but are not a second product.

Phase coverage

• Beta1: released usability, Official continuation, provider-scoped context window, catalog, release, and Hosted-CI regressions.
• Beta2: generic tool compatibility core, exact identity, four dispositions, and no cross-Provider proxy.
• Beta3: generic client-executed tool_search and reversible Code Mode workflows.
• Beta4: generic Collaboration V2 lifecycle plus V1/configuration isolation.
• Beta5: full independent Responses regressions for the Provider configurations maintained by CodexHub.
• Beta6: runtime fields, progressive Chat, multi-turn text, standard tools, reasoning defaults, and read/apply_patch/verify.
• Beta7: Provider/persistence/availability/retry/teardown and bounded-evidence regressions.

Profile requirements

• Bind exact candidate SHA, normalized Codex CLI version, profile/catalog hashes, Provider, canonical model, inbound/upstream protocol, case contract, and terminal result.
• Use GLM-5.2 and K2.7 Code as maintained-Provider regression samples, not as a runtime eligibility list.
• Add cold startup, profile regeneration/readback, same-Home restart, reconnect, clean cancellation, and Beta1 Official-switch controls.
• Isolate each case's Codex home, workspace, config, state, processes, and transient artifacts.
• No Desktop, ZCode, OpenCode, Pi, OMP, or manual GUI interaction is required by this automated profile.

Acceptance criteria

• The profile is declarative and versioned.
• Fixtures cover generic scenario binding, timeout, cancellation, cleanup, and failure classification.
• Real CLI cases prove client completion plus correlated exact-route evidence.
• Below-floor, ambiguous, or unparseable CLI versions cannot count as accepted release evidence.
• A newer unknown runtime shape is reported as an unrecognized verification profile without changing product model eligibility.
• No hidden case retry or selective replacement occurs; every attempt remains visible.
• Cleanup runs and is reported on success, failure, timeout, and cancellation.
• Durable evidence satisfies Harden bounded real-client capture and preserve the accepted 0.1.7 profile #244/0.1.8 RC1: Aggregate six fixed regression batches and publish promotion evidence #257 bounds and contains no raw secrets or unbounded work roots.

Blocked by

• Harden bounded real-client capture and preserve the accepted 0.1.7 profile #244