Context Pipeline

Unified two-stage context collection and curation, including structured native task context and separate Obsidian vault context.

Details

Two-stage pipeline. ContextCollector fetches raw JSON per source with cache awareness (max_age + source-level is_stale). ContextCurator renders any cached Context into markdown or JSON at the caller's depth. Structured sources include git, native tasks, projects, and chrome; vault-oriented wrappers remain separate. Exposed over MCP as context_block + context_drill_down.

After native task activation, TasksSource and the task section of the Obsidian collector query TaskStore. They never scan the frozen master list or task-note Markdown as current task truth.

GitSource is multi-repo since the Phase-A migration: it walks every .git directory at depth 1 under cfg['repos_root'], tags commits with a per-repo project field, and renders them bucketed under #### <project> subheadings. Pass custom={'git': {'repo_path': ...}} to force single-repo scope. The legacy work_buddy/collectors/git_collector.py is retained for test fixtures and historical callers but is no longer on the bundle path.

Project sync lives at work_buddy/projects/sync.py (formerly work_buddy/collectors/project_collector.py, which is now a back-compat shim). Entry: sync_projects(cfg) -> str (alias: collect). This is a synthesis job that writes the SQLite registry, not a context fetcher — ProjectsSource reads the already-synced registry.