| name | kg-kvcache |
| skill_type | skill |
| description | Thin verb over the engine's shared, content-addressed KV-cache — get/put opaque blocks, check membership, and read occupancy/dedup stats. Use for the shared block cache — "cache this block", "is this key cached", "kvcache stats". |
| license | MIT |
| tags | ["graph-os","engine","kvcache"] |
| tier | core |
| metadata | {"author":"Genius","version":"0.1.0"} |
kg-kvcache
Condensed intent-surface note (Seam 8). Under the default intent surface (MCP_TOOL_MODE=intent), graph_kvcache is held back from the default tool list (nothing removed — REST + _execute_tool still reach it exactly as documented below). Two ways to use this skill unchanged: (1) load_tools(tools=["graph_kvcache"]) once per session (as below), then proceed exactly as documented; or (2) call the manage intent verb with the same natural-language request — the resolver routes to graph_kvcache for you and returns the result plus a routing justification. Set MCP_TOOL_MODE=condensed/verbose/both to expose the granular tools eagerly instead.
graph_kvcache (CONCEPT:AU-KG.coordination.engine-message-broker, via the KG-2.306 EpistemicGraphKVBackend) is the engine's content-addressed KV-cache. Actions: get (key→base64 block bytes or miss), put (key+value_b64→stored bool), contains/exists (key→bool), stats (occupancy + dedup counters). The connector degrades every transport error to a cache miss, so it never raises.
Invoke
- MCP:
load_tools(tools=["graph_kvcache"]), then graph_kvcache(action="stats").
- REST twin:
POST /graph/kvcache with {"action": "get", "key": "..."}.
Example
graph_kvcache(action="put", key="blk:abc", value_b64="...")
