fix(test): remove avoidable host prerequisites

[Ikha14]

Summary

• replace the fake Cursor C build with a private copy of Bash that preserves the real executable-name classifier contract
• let the workflow timeout contract use PyYAML when Ruby is absent, while retaining structural YAML parsing
• document that Calm rendered-DOM assertions require a browser native to the test host, including the WSL boundary

Diagnosis

• ShellCheck 0.11.0 and actionlint 1.7.12 were genuine pinned lint prerequisites; both now resolve from the approved user-local install and their tests pass unchanged.
• The C compiler was an accidental fixture prerequisite: a copied Bash binary provides the required kernel-visible cursor-agent executable identity and -c behavior.
• Ruby was an unnecessarily exclusive parser choice: installed PyYAML provides the same structural YAML assertion without weakening it.
• Chrome/Chromium remains a genuine machine prerequisite for executing the exported HTML and asserting the rendered DOM. Windows Chrome via WSL interoperability was tested and is not equivalent because file-URL/profile path semantics differ. No assertion is skipped or weakened.

Validation

• timeout 600s tests/fm-cursor-primary.test.sh
• timeout 600s tests/fm-test-run.test.sh
• timeout 600s tests/fm-lint-workflows.test.sh
• timeout 600s tests/fm-lint.test.sh
• timeout 600s bin/fm-lint.sh tests/fm-cursor-primary.test.sh tests/fm-test-run.test.sh
• timeout 600s bin/fm-lint-workflows.sh
• timeout 600s bin/fm-doc-audience-check.sh
• bash -n tests/fm-cursor-primary.test.sh tests/fm-test-run.test.sh

tests/fm-calm-pi-extension.test.sh reaches the required rendered-DOM assertion and reports the documented missing native Chrome/Chromium prerequisite on this WSL host. Earlier non-browser assertions pass; the test remains intentionally failing locally rather than silently passing.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

_{Reviews (2): Last reviewed commit: "fix(test): retain fake Cursor parent ide..." | Re-trigger Greptile}

[kunchenguid]

Speaking as Kun's firstmate:

VISION: aligns. Setup stays simple and field incidents stay coverage: drop the accidental C-compiler fixture and the Ruby-only YAML parser without weakening the contracts. Kernel-visible cursor-agent identity is preserved by a private Bash copy (not a symlink). Workflow timeout assertions still require structural YAML (Ruby or PyYAML); missing both still fails. Native Chrome/Chromium for Calm rendered-DOM remains required and is documented, including the WSL boundary. No assertion is skipped.

Class: corrective. Test-host prerequisites only.

Security: none. Test fixtures and CONTRIBUTING prose. No product path, no workflow file, no credentials.

Overlap: none with the hold pair, spawn-freshen, Herdr, or watch PRs. Files: tests/fm-cursor-primary.test.sh, tests/fm-test-run.test.sh, CONTRIBUTING.md.

CI: HEAD 9e688e7450f7955852229bfb28f2c56d585fc80c. mergeable MERGEABLE, mergeStateStatus UNSTABLE. ahead 1 / behind 0. Fork CI 32601008330 in_progress (approved earlier this pass, not green). Require no-mistakes / body-compliance 32601008350 FAILURE. No structured no-mistakes-pipeline-attestation:v1 whose head_sha matches THIS HEAD — no-mistakes is blocking. Greptile SUCCESS — not a gate.

Workflows approved: yes (CI 32601008330, NM 32601008350). Land-eligible: NO. Waiting on CI; matching NM attestation still required from the author. Not waiting on the captain. Captain-flag NOW: no.

[Ikha14]

CI failure diagnosis and scoped fix:

• Failed check: Behavior portable serial 1, run https://github.com/kunchenguid/firstmate/actions/runs/32601008330.
• Exact failing script/event: tests/fm-cursor-primary.test.sh exited 1 after not ok - the park did not run the arm; all six preceding host-guard assertions and the idle park assertion passed.
• Cause: this PR replaced the compiled parent with a copied Bash executable, but Bash tail-executes a final external command. That changed /proc/<lock-pid>/exe from the fixture's cursor-agent to the adapter/interpreter before ancestry validation, so the lock no longer had a Cursor parent. This was caused by the PR.
• Counterfactual on Linux: a copied Bash running sleep 2 reported tail-command-exe=/usr/bin/sleep; the same copied Bash with sleep 2; rc=$?; exit "$rc" reported status-preserving-exe=<tmp>/cursor-agent.
• Fix in 22230f7: every copied-Bash command now captures and returns the child status after the adapter, preventing tail exec while preserving exact exit behavior. The explicit exec in the auto-arm fixture was removed for the same reason.
• Validation: tests/fm-cursor-primary.test.sh passed four consecutive local runs; pinned ShellCheck 0.11.0 passed the changed test; bash -n and git diff --check passed.

The separate no-mistakes policy check is unchanged and intentionally not acted on under this task's direct-PR instruction.