Phase 3: screen capture + indexing

[Copilot]

Implements Module 3 from the agent plan: periodic multi-monitor screenshot capture, OCR, LLM summarisation, privacy filtering, and SQLite indexing — plus the snap one-shot CLI command.

New packages

internal/store

Minimal SQLite store backing all future phases. Runs schema migrations on Open. InsertEvent is the write path for capture (and future) events.

internal/capture

Full capture pipeline assembled as a Capturer struct:

CaptureAll()  →  OCR()  →  PrivacyFilter.MatchesText()  →  Summarise()  →  IndexCapture()

• capture.go — CaptureAll via kbinani/screenshot; OCR shells out to tesseract (degrades gracefully to empty string when not installed); Summarise calls the engine with "Summarise the following screen content in 2 sentences: …"
• privacy.go — loads ~/.super-ollama/privacy.toml (XDG-aware); filters on skip_keywords (case-insensitive text match) and skip_window_titles (compiled as case-insensitive regex patterns)
• scheduler.go — StartScheduler wraps robfig/cron at the configured interval; returns a draining stop() func

Updated

• internal/config — adds capture_interval_minutes (default 5) and capture_enabled
• cmd/super-ollama/main.go — replaces snap stub with a real command: loads config + privacy filter + engine + store, runs Snap(), prints per-display summaries

New dependencies

• github.com/robfig/cron/v3 v3.0.1
• github.com/kbinani/screenshot v0.0.0-20230812210009-b87d31814237