fix(test): support Windows virtualenvs in canonical runner

[lucamoretti09]

What

Makes the canonical scripts/run_tests.sh work with both virtualenv layouts:

• POSIX: .venv/bin/python
• native Windows: .venv/Scripts/python.exe

It also checks both .venv and venv under the repository and installed Hermes checkout, and preserves USERPROFILE plus PYTHONUTF8=1 inside the runner's hermetic env -i environment.

The bundled Hermes skill documentation is updated to distinguish a supported Windows venv layout from the smaller end-user venv that may intentionally omit pytest.

Verification

• 2 passed — tests/scripts/test_run_tests_windows_venv.py
• bash -n scripts/run_tests.sh using Git for Windows Bash
• git diff --check
• staged secret-pattern scan: 0 hits

Scope

This only changes interpreter discovery and the clean environment. POSIX behavior and test-runner arguments remain unchanged.

[teknium1]

Thanks for addressing native Windows test-runner discovery. The underlying defect is present on current main: scripts/run_tests.sh:42-54 only accepts bin/activate and derives bin/python, so Scripts/python.exe venvs cannot run.

Problems

• tests/scripts/test_run_tests_windows_venv.py:10 reads run_tests.sh and asserts source strings. This is a banned source-snapshot pattern under AGENTS.md:1358-1411; it does not exercise interpreter resolution or the hermetic environment.
• website/docs/user-guide/skills/bundled/autonomous-ai-agents/autonomous-ai-agents-hermes-agent.md:7 says the page is generated from skills/autonomous-ai-agents/hermes-agent/SKILL.md. The PR edits only the generated page, while the source skill still says the wrapper is POSIX-only at skills/autonomous-ai-agents/hermes-agent/SKILL.md:868 and :1067.

Suggested changes

• Replace the snapshot test with a behavioral runner test using a temporary fixture and real shell invocation; cover both interpreter layouts.
• Edit the source SKILL.md and regenerate its website page.

This is an automated hermes-sweeper review.

[teknium1]

This test snapshots shell source rather than executing behavior, which AGENTS.md:1358-1411 bans. Please replace it with a behavioral invocation against a temporary fixture/venv so it proves the selected interpreter and clean environment work.

[teknium1]

This page is generated from skills/autonomous-ai-agents/hermes-agent/SKILL.md (its header says so). Please update that source skill and regenerate this page; otherwise the shipped skill continues to say the wrapper is POSIX-only.