chore: triage D01, D10 and the [BLOCKED] hook exit-code finding

[pacphi]

Why

The v5 review sent these to v4 for triage under Decision 4 (ledger G8).

In scope

Reproduce each on 4.0.0-alpha.60; give each a disposition and, when exit-critical, a card.

Out of scope

Fixing them here.

Acceptance criteria

• Each item has a recorded disposition
• Exit-critical items have their own alpha.61 card

Dependencies

Blocked by: None

Blocks: #320

Plan

• Release: 4.0.0-alpha.61
• Size: size:S · Workstream: ws:remediation
• Epic: chore: exit release 4.0.0-alpha.61 (epic) #294
• Card: A61-16 in the card inventory
• Sources: Supersession ledger (docs(plans): v4 GA master plan, supersession ledger and v5 registers #290) G8; v5 impact review (docs(plans): v4 GA master plan, supersession ledger and v5 registers #290) Items that belong on the v4 board

Verification

Each acceptance criterion is checked on the pull request.

[pacphi]

Triage, 2026-10-02, with the maintainer's approval of each disposition (option A on all three).

"[BLOCKED]" hook exit-code finding (Superpowers vs. the rUv Stack audit): reproduced, upstream-fixed, not exit-critical.

• Ruflo's pre-bash helper prints [BLOCKED] Dangerous command detected: rm -rf / and then exits 1. Claude Code treats exit 1 as a non-blocking error; only exit 2 blocks. Ruflo's own code comment says so: "exit 2 blocks execution; exit 1 is non-blocking".
• I ran the helper from each published @claude-flow/cli release on a dangerous command. Releases 3.5.19 through 3.51.0 (every one sampled, including 3.47.0, the version on the maintainer's machine) exit 1. 3.51.1, published 2026-10-02 16:10Z, exits 2. The diff between 3.51.0 and 3.51.1 is that one line plus its comment.
• A project's copy of the helper, written by an older ruflo init, keeps the old behaviour. ak's own OpenCode template blocks on the [BLOCKED] text, so only the Claude Code hook path was exposed.
• Disposition: not exit-critical for alpha.61. Carried as beta.1 card test: flag project hook helpers whose pre-bash exits 1 on [BLOCKED] #428 (ak audit hooks diagnostic for a helper that exits 1 after [BLOCKED]). Raising the Ruflo support floor is decided separately.

D10 (an unexplained "31MX" fragment in terminal output): closed as not reproducible without a capture. No raw capture was available. Reopen if one turns up.

D01 (Usage, History and Models stay empty until the browser is refreshed): still open, waiting on one check on the maintainer's machine. I could not reproduce it: in a real browser against fixture transcripts (main), a cold load and a reload were identical on Usage, Observability, Overview and System, and Usage showed its sessions on first load. The report probably depends on the size of a real transcript corpus, so the maintainer will open the dashboard cold once on the Mac and click Usage. If it shows empty until reload, this becomes a defect with a card; if not, it closes as not reproducible with a first-load regression test carried into P6.

This card closes when D01 has a recorded disposition.

---

Generated by Claude Code

[pacphi]

D01 now has its disposition, so every item on this card does.

D01 (Usage, History and Models stay empty until the browser is refreshed): not reproducible as stated; carried into P6 as a first-load requirement.

• On 2026-10-02 the maintainer opened the dashboard cold on 4.0.0-alpha.60 (ak --version), on the Mac with a real corpus: 1,034 sessions in the 14-day window. The first screen took almost 10 s to appear. Usage took a little over 5 s and Observability a little over 5 s. Nothing was empty once it appeared, and the screenshots show Usage (sessions, cost, tokens, cache and the cost-per-day chart) and Observability (Live) fully populated. The maintainer did not need a reload.
• Earlier I could not reproduce it either: in a real browser against fixture transcripts, cold load and reload were identical on every primary tab.
• The likely source of the report is the wait itself. With a large corpus a panel can read as empty for 5 to 10 s before its data arrives, and a reload then feels like the fix because the server's cache is warm. This is a reading of the evidence, not a confirmed cause.
• Disposition: not a defect to fix in alpha.61. P6 rewrites the Usage readers, so the requirement goes to P6-12 (feat: Usage and Observability follow the scope filter #374): on a corpus of about 1,000 sessions, no Usage or Observability panel reads as empty while its data loads, and a test loads each panel cold.

Result for the card

• "[BLOCKED]" finding: reproduced, upstream-fixed in 3.51.1, not exit-critical; beta.1 card test: flag project hook helpers whose pre-bash exits 1 on [BLOCKED] #428.
• D10: closed as not reproducible without a capture.
• D01: not reproducible; first-load requirement carried to P6-12.
• Nothing is exit-critical, so no alpha.61 card is needed. Both acceptance lines are met.

---

Generated by Claude Code