Phase 6: ~/.super-ollama/ layout, --no-profile flag, install.sh, build matrix

[Copilot]

Implements the full Phase 6 scope: file-based profile/prompt configuration under ~/.super-ollama/, profile injection into every inference call with an opt-out flag, a cross-platform install script, and a CI build matrix.

internal/config

• Extended Config struct with all config.toml fields: EmbeddingModel, DBPath, TodoPath, CaptureIntervalMinutes, CaptureEnabled
• Added Dir() — resolves ~/.super-ollama/ (honours XDG_CONFIG_HOME)
• Added LoadProfile() — reads profile.md; silent no-op when absent
• Added LoadPrompt(name) — reads prompts/<name>.md; silent no-op when absent
• Added Init() — creates the directory skeleton with default files; idempotent

internal/engine

• Generate and StreamGenerate now accept a system string param forwarded to api.GenerateRequest.System

cmd/super-ollama

• Added --no-profile persistent flag (suppresses profile.md injection on all commands)
• ask: injects profile.md + prompts/ask.md as the system prompt
• chat: prepends profile as a system message before the conversation loop
• Added config init subcommand; config show now displays embedding_model, directory path, and profile status

super-ollama --no-profile ask "summarise this"   # skip profile injection
super-ollama --model llama3:8b ask "hello"       # override model per-call
super-ollama config init                          # scaffold ~/.super-ollama/

install.sh

One-liner cross-platform installer: detects OS/arch, downloads the release binary, and runs Init() equivalent to scaffold ~/.super-ollama/.

.github/workflows/build-matrix.yml

Builds super-ollama for linux/{amd64,arm64}, darwin/{amd64,arm64}, and windows/amd64. Separate test-prompt-pipeline job runs internal/config and cmd/config tests on all three OSes.