Context Pipeline

Unified two-stage context collection + curation. ContextCollector fetches raw JSON from registered sources (git, tasks, projects, chrome + 9 markdown wrappers), ContextCurator renders into depth-adapted markdown or JSON. Feeds LLM prompts (build_triage_context retrofits onto this) and bundle files (collect.py retrofits onto this).

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. 13 registered sources split into structured (git/tasks/projects/chrome with drill-down) and markdown-wrapper (9 sources delegating to legacy collectors). Exposed over MCP as context_block + context_drill_down.

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.