Install
openclaw skills install uncagdValidated CAG decision memory for OpenClaw tools, agents, and long-running project sessions.
openclaw skills install uncagdUse this skill when the user is working on a repo, software project, agent workflow, plugin, MCP server, or multi-step implementation where decisions may matter later.
Before planning or editing code, call cag.retrieve when:
Use k=5 unless the user explicitly asks for a broader review.
Set useEmbeddings=true only as a secondary ranker when lexical retrieval is weak.
Call cag.capture_candidate when the conversation produces a durable decision such as:
Captured candidates are not durable until validated.
Call cag.validate_memory only when:
Do not store raw assistant guesses as validated memory. Do not use CAG memory as a replacement for reading the codebase. Use retrieved decisions as constraints, not as proof of current code state.
When contradictions are detected:
cag.resolve_contradiction with keep_existing if the prior decision stands.supersede_existing if policy changed and old validated memories should be deprecated.create_candidate if the new decision needs review first.cag.compress_memory to consolidate old validated decisions into one compact, validated summary.cag.export_memory with format=memory_md for portable handoff docs.cag.import_memory to rehydrate memory in another workspace.