fix(workflows): init step must not replace init's own error with SystemExit: 1 - #4530

Problem

InitStep._run_init appends the runner's exception to stderr:

if result.exit_code != 0 and result.exception is not None:
    detail = f"{type(result.exception).__name__}: {result.exception}"
    stderr = f"{stderr}\n{detail}".strip() if stderr else detail

That branch exists for an unexpected crash. But typer.Exit(n) — how specify init reports every ordinary failure — surfaces through CliRunner as result.exception = SystemExit(n), so it fires on routine errors too.

init prints its diagnostics through Rich to stdout, so result.stderr is empty. The synthesized detail therefore becomes the entire stderr, which preempts execute's fallback two frames later:

error=(
    stderr.strip()
    or stdout.strip()          # <-- never reached
    or f"specify init exited with code {exit_code}."
),

Reproduction on current main (c173bf1)

An ordinary typo in integration: — which InitStep.validate does not value-check:

validate     : []
status       : failed | exit_code: 1
result.error : 'SystemExit: 1'
output.stderr: 'SystemExit: 1'
stdout holds : "... lingma, muse, omp, opencode, pi, qodercli, qwen,
                rovodev, shai, tabnine, trae, vibe, zcode, zed"

So the workflow author is told SystemExit: 1 while the list of valid integrations sits unread in stdout. The same applies to any ordinary init failure — a project: directory that already exists, an unusable script type.

It also leaks into workflow data: steps.<id>.output.stderr is the string "SystemExit: 1", so a downstream step reading it gets the sentinel rather than a diagnosis.

Fix

Exclude only SystemExit, preserving the branch for genuine crashes:

if (
    result.exit_code != 0
    and result.exception is not None
    and not isinstance(result.exception, SystemExit)
):

After the fix the same input reports init's own message, and a RuntimeError escaping the runner still yields RuntimeError: boom inside init.

Verification

• Fail-before / pass-after: 1 new-vs-baseline failure with the source reverted to upstream/main → 16 passed with the fix.
• A second test pins the branch's original purpose — an unexpected RuntimeError still surfaces. It passes both before and after by design: it guards preserved behaviour rather than proving the fix.
• Scoped regression on tests/test_workflows.py: 20 failed / 959 passed vs a clean-main baseline of 20 failed / 957 passed — no new failures (the 20 are the known Windows os.replace flakiness in that file).
• uvx ruff@0.15.0 check src tests → clean

Behaviour change, disclosed: error and output.stderr for a failing init step change from "SystemExit: 1" to init's own captured output (which includes its banner, since init writes to stdout). Successful steps are untouched, and no existing test asserted the old sentinel.

Written with assistance from Claude Code. Bug found, reproduced, and verified by me on current main.

🤖 Generated with Claude Code