McRitchie Studio

← All Tasks

Sizing

Harden release_cli subprocess capture

Archived

harden-release-cli-subprocess-capture

Created

Jun 25, 05:54

Started

Jun 25, 12:42

Completed

Jun 25, 19:24

DevOps handoff

Type

Bug

Shape

backend

Worktree Slug

harden-release-cli-subprocess-capture

Repositories

mcritchie-studio

Release Train

—

Branch

feat/harden-release-cli-subprocess-capture

Pull Request

https://github.com/amcritchie/mcritchie-studio/pull/193

Local URL

—

QA URL

—

Production URL

—

tooling ci-flake

Acceptance Criteria

release_cli_test IO.popen helpers capture exit status
Flunk on subprocess failure not empty stdout
Preserve line-126 legitimate empty-output assertion

Expected Test Plan

[unit] helpers loud-fail on subprocess error

Checks Run

[unit] release_cli_test 70/70 standalone; 50/50 tight loop (~3500 subprocess runs) green
[unit] regression fail-first proven: RED vs old IO.popen helper, GREEN vs hardened run_ruby (stderr+exit=1 surfaced)
[unit] line-126 legitimate empty-output canary still passes (no over-guard) + new empty-on-clean-exit canary
[integration] PARALLEL_WORKERS=4 bin/rails test release_cli_test 6/6 green (parallel-fork harness, the flake's real I/O boundary)

Agent Context

Follow-up to stabilize-worktree-port-test (PR #186). Steffon found test/lib/release_cli_test.rb carries the IDENTICAL latent flake: IO.popen(..., {err: File::NULL}, &:read) across 3 helpers (eval_helper / eval_with_argv / run_cli) discards stderr + exit status, so a reaped-before-flush child surfaces as a silent empty-output failure under CI's parallel-fork harness — same mode as agent_worktree_test:76. It's also in the gating CI sweep. NOT folded into #186 (different file, larger review surface) AND it needs a GENTLER guard: line 126 (test_unparsed_flag_returns_nil_through_the_bin_boundary) LEGITIMATELY asserts empty output, so the helper here must flunk on NONZERO EXIT only, not on empty stdout (unlike the worktree test where all 5 checks return non-empty and the stronger flunk-on-empty guard was safe). Apply Open3.capture3 + exit-status check, preserving the legit empty-output case.

Stage Timeline

Who handled each stage, the time it took (measured), and the model / tokens / cost reported (best-effort) — plus who's on it right now. — means the agent didn't report that metric.

Created → Designed

N
Nidoran♀

Model

—

Duration

—

Tokens

—

Cost

—

Completed Jun 25, 05:54 · 2 days ago

api
Designed → Building

N
Nidoran♀

Model

claude-opus-4-8

Duration

about 7 hours

Tokens

—

Cost

—

Started Jun 25, 05:54

Completed Jun 25, 12:42 · 2 days ago

cli
Building → Submitted

N
Nidoran♀

Model

claude-opus-4-8

Duration

12 minutes

Tokens

—

Cost

—

Started Jun 25, 12:42

Completed Jun 25, 12:54 · 2 days ago

cli
Submitted → Reviewed

C
Carl primary

S
Shannon light

Model

claude-opus-4-8

Duration

about 1 hour

Tokens

3,368,320

Cost

~$6.19

Started Jun 25, 12:54

Completed Jun 25, 13:52 · 2 days ago

cli
Reviewed → Assembled

S
Steffon

Model

—

Duration

about 5 hours

Tokens

—

Cost

—

Started Jun 25, 13:52

Completed Jun 25, 19:03 · 1 day ago
Assembled → Shipped

A
Avi

Model

—

Duration

21 minutes

Tokens

—

Cost

—

Started Jun 25, 19:03

Completed Jun 25, 19:24 · 1 day ago
Shipped → Archived

8
81fad33e-98b5-4075-8ed6-d0efe6cb5804

Model

claude-opus-4-8

Duration

about 2 hours

Tokens

—

Cost

—

Started Jun 25, 19:24

Completed Jun 25, 21:16 · 1 day ago

cli