knowledge-layer

# Knowledge Layer Skill ## Purpose Provide a high-level deployment wrapper over RepoBrain core, with graph-first knowledge injection and all-file support (code, docs, data, media metadata). ## Inputs - `refresh_filesystem(workspace=".", full=False)` - `ask_filesystem(question, workspace=".")` Refresh automatically chooses a first complete build, matching interrupted-task resume, committed-diff update, or no-op. `full=True` always starts a new complete build. The workspace defaults to the current directory and must be a clean Git worktree. Ask stays read-only and only suggests `rb-refresh` when knowledge is stale. Both refresh surfaces report failures without replacing usable knowledge. ## Outputs - Refresh writes graph-first artifacts under `.repobrain/`: - `knowledge_graph.json` - `knowledge_graph.md` - `knowledge_graph.mmd` - `document_index.md` - `data_overview.md` - `media_manifest.md` - plus existing `conventions.md` and `structure.md` - Ask returns a grounded answer with source paths. ## Boundaries - Skill is a wrapper layer only; no standalone runtime. - Core Hub/Agent/Pipeline architecture remains the source of truth. ## Compatibility - Existing commands (`rb-refresh`, `rb-ask`, `rb-mcp`) remain valid. - High-level aliases can be disabled with `RB_ENABLE_LAYER_ALIASES=0`. ## Degrade Strategy - If graph artifacts are unavailable, ask falls back to `structure.md` and `conventions.md` context.