fix(lifecycle): stop install fallback when cancellation races managed install

[cursor]

Summary

Fix a high-severity lifecycle bug where a cancelled managed install could fall through to the next install method without rolling back a package that had already been installed on disk.

When waitForSpawnedCommand() reports failure because the CLI context is cancelled after a child exits 0, installAgent() now rolls back the current managed method and stops instead of trying later install methods (for example npm after bun).

Linked Artifacts

• Issue:
• ADR:
• OpenSpec: openspec/changes/fix-install-cancel-method-fallback
• Discussion:

Validation

• bun run memory:check
• bun run lint
• bun run format:check
• bun run typecheck
• bun run test (if behavior changed)
• Not run, explained below

Release Intent

• Release: patch - bug fix

Docs Updated

• Not needed
• docs/...
• openspec/...
• Follow-up issue or OpenSpec change created instead

Scope Check

• I did not add a new ad hoc root-level Markdown file.
• I updated the relevant issue, ADR, spec, runbook, or captured the missing doc work as follow-up.
• I did not silently expand project scope without recording it explicitly.

Closure Check

• Working tree was clean after commit.
• Branch was pushed and this PR is the active delivery artifact.
• OpenSpec change is not needed, still active by design until merge, already archived, or queued for agent-driven archive closure.
• Release is not applicable, delegated to release automation, or verified.

Notes

Bug impact: Duplicate global installs and orphaned packages when install is cancelled by timeout or signal after the first managed method succeeds on disk.

Trigger scenario: quantex install <agent> --timeout … or exec/shortcut install with timeout; bun (or other managed PM) completes add -g with exit 0, then cancellation fires before Quantex records success.

Root cause: waitForSpawnedCommand() returns exit code 1 on cancellation even when the child exited 0. installAgent() treated that as a failed method attempt and continued to the next install method without rollback.

Fix: Guard each install-method iteration for cancellation; when executeMethod() returns false and the context is cancelled, roll back the current method and return failure.

Remaining owner: Maintainer merge + release-please patch release; archive follow-up for fix-install-cancel-method-fallback after merge.

  

[cursor]

PR Governance — #405

Verdict: Non-blocking governance pass. No approval. PR already reached merge delivery (79a26d4, merged by @Drswith).

OpenSpec intake

• Correct for observable lifecycle behavior change: fix-install-cancel-method-fallback created with proposal/design/spec delta/tasks before src/ edits.
• Change correctly remains active until archive closure; Notes document the archive follow-up owner.

Validation

• Checklist matches the changed surface: memory:check, lint, format, typecheck, and test for src/package-manager/ behavior.
• CI classified and ran the full matrix + sandbox-tests; all checks green. No evidence of blind CI patching.

PR body / closure

• Template sections present; validate-body passed in PR Governance workflow.
• Release intent patch is appropriate for fix(lifecycle):.
• Closure check correctly marks OpenSpec active-by-design and delegates release to automation.

Non-blocking follow-ups

1. Archive closure: Spec delta re-lists existing Windows/batch cancellation scenarios. Archive agent must merge additively into openspec/specs/agent-update/spec.md (add only the new “Cancelled managed install does not fall through…” scenario).
2. Traceability: No linked GitHub issue for the regression context.
3. Checklist gap: bun run openspec:validate is in tasks but not the PR Validation checklist.

Remaining owners

• OpenSpec archive closure: archive follow-up PR for fix-install-cancel-method-fallback
• Release closure: release-please patch from merged fix: commit

No reviewer request — maintainer-authored commit and self-merge.

  

_{Sent by Cursor Automation: Quantex CLI PR Governance}

[cursor]

Non-blocking (archive): This delta re-lists existing Windows wrapper and batch cancellation scenarios. Archive closure must merge additively into openspec/specs/agent-update/spec.md — add only the new scenario at lines 32–40, do not replace or drop the existing requirement text.

[cursor]

Cancellation guard looks correct: loop-start check avoids work when already cancelled; post-executeMethod(false) rollback only runs when cancelled is set, so genuine method failures still fall through to later methods.