gate Ollama install/start/pull/launch behind y/N prompts

[StewAlexander-com]

Summary

Make ./install.sh (and ./run.sh) detect missing prerequisites and prompt the user y/N before doing anything that touches the host. The default answer is no, so a stray Enter never installs a system binary. This keeps the "minimal intervention" promise of the two-script flow while removing the prior hard requirement that the user install/start Ollama by hand.

• install.sh — adds a confirm() helper and gates four actions behind y/N:
  • install Ollama (runs brew install ollama on macOS or the upstream curl ... | sh on Linux),
  • start ollama serve in the background (nohup, logged to /tmp/ollama-serve.log),
  • ollama pull "$TUTOR_MODEL" (default gemma3:4b),
  • exec ./run.sh to launch the app immediately after setup.
• run.sh — mirrors the helper and offers to start ollama serve when the binary is present but the daemon is not. Bootstrapping a fresh venv via install.sh is now done in noninteractive + skip-Ollama mode so run.sh never installs system binaries silently.
• Env contract — accepted yes responses: y, Y, yes, Yes, YES. All four prompts default to "no" otherwise.
  • TUTOR_NONINTERACTIVE=1 / PYTHON_TUTOR_NONINTERACTIVE=1 → auto-no
  • PYTHON_TUTOR_ASSUME_YES=1 → auto-yes (suitable for pre-approved unattended setup)
  • PYTHON_TUTOR_AUTOLAUNCH=1 → exec ./run.sh after install without asking
  • existing TUTOR_SKIP_OLLAMA=1 / TUTOR_SKIP_MODEL_PULL=1 continue to work
• Docs — README quick start now has an "Opt-in install prompts" table and a non-interactive/CI usage block. docs/install-runtime-workflow.md gets a refreshed Decision section and a new mermaid flow diagram showing every y/N gate.
• Tests — new scripts/smoke_prompts.sh exercises the noninteractive path, the PYTHON_TUTOR_NONINTERACTIVE alias, and the PYTHON_TUTOR_ASSUME_YES auto-launch (via a temporarily symlinked run.sh stub so CI doesn't actually start uvicorn). Wired into the existing scripts CI job.

Behavior

Default flow (interactive, no Ollama present on a Linux box):

[install] using python3.12 (Python 3.12.8)
[install] venv already present at backend/.venv
[install] upgrading pip and installing backend deps
[install] backend dependencies installed
[install] Ollama is not installed.
[install] Install Ollama now? (will run the official upstream installer) [y/N] _

Pressing Enter (no) prints the manual command and moves on. Typing y runs curl -fsSL https://ollama.com/install.sh | sh. The next prompts (Start ollama serve…, Pull gemma3:4b…, Launch the tutor now…) are gated identically.

Testing

Local, on this branch:

• bash -n install.sh run.sh scripts/smoke_prompts.sh — syntax check.
• TUTOR_SKIP_OLLAMA=1 TUTOR_NONINTERACTIVE=1 ./install.sh — completes, prints [auto-no] on the launch prompt, leaves backend/.venv usable.
• PYTHON_TUTOR_NONINTERACTIVE=1 TUTOR_SKIP_OLLAMA=1 ./install.sh — alias works identically.
• PYTHON_TUTOR_ASSUME_YES=1 TUTOR_SKIP_OLLAMA=1 ./install.sh (with run.sh symlinked to a stub) — prints [auto-yes], execs ./run.sh.
• ./scripts/smoke_prompts.sh — all three runs green.
• ./scripts/smoke_run.sh — server boots, /api/health returns degraded (no Ollama), / serves the SPA.
• backend/.venv/bin/pytest -q — 89 passed in 3.06s.

CI runs scripts/smoke_prompts.sh alongside the existing install + run smoke jobs.

---

🤖 Generated by Computer