test(braid): capture installed CLI setup recovery proof

[drewstone]

Change

• Extend the installed CLI proof through setup save, connection A→B→A, explicit task submission, encrypted disk reopen, and same-branch continuation refusal.
• Assert authenticated preflight ordering, bounded idempotent retries, credential redaction, retained/cleared rejected draft, /new transcript, and temporary-state cleanup.
• Keep task outcomes unknown when loopback writes return HTTP 503. The fixture creates no cloud tasks or resources.
• Add uncut terminal captures and labeled 4x copies for 40×12, 80×24, 120×40, and 200×60, plus the media manifest and maintained release record.

Evidence

• Exact installed candidate: source 3b627e9199186ccfb523f07d8c52effe02e46281, Release run 36562109729, archive SHA-256 f13e3bea3bbdda567cc125cde9cbbe1889625e0972fa2303da17210c47e102d9, Node v22.23.2.
• Four dimensions passed. Each run captured 26 loopback requests; both tasks reached five bounded create attempts and remained unknown; encrypted reopen preserved the run and message; same-branch refusal made read-only requests only; cleanup and credential assertions passed.
• Harness PIDs, CLI PIDs, request evidence, per-run logs, and recording markers are in media README and manifest.
• Biome and Node syntax checks passed. Normal pre-commit and pre-push hooks passed. Luna's final independent review found no concrete helper or recording findings.

Release boundary

This PR proves installed CLI and local storage behavior with a loopback fixture. It does not prove live Tangle execution or provider behavior, and it does not publish or deploy Braid 0.3.4. Public npm and GitHub release remain 0.3.3.

The protected GLM-5.3 attempt remains RUNTIME_FINAL_ERROR with no retained provider status/body, generation ID, usage, cost, or cleanup census. Its cause and cleanup result remain unknown. The protected live workflow has no preflight-only mode, and no already-authorized bounded provider path is identified. Keep publication held until exact-candidate protected live evidence and required cleanup proof pass.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[tangletools]

✅ Auto-approved PR — 5678bf5e

Blanket team auto-approval is intentional. The merge gates are CI and codex-p1.
No automated review runs on this PR. This approval rests on the rule above alone.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-09-30T02:20:03Z}