Hi! I’d appreciate advice on how to structure this in OpenClaw.
I’m building a personal OpenClaw setup, and my immediate goal is practical: map the current setup and maintain a “setup inventory” over time.
Background: this started from Karpathy’s point about LLM-native workflows / “soft programs” — .md skills or playbooks are not just docs or shell-script wrappers, but instructions an agent can execute with judgment: context, permissions, fallbacks, verification, rollback, etc.
What I want first is not a big skill library, but a reliable current-state map:
- agent capabilities
- permissions and boundaries
- channels/integrations/models
- workspace/project structure
- setup changes over time
- what should be updated next
- rollback path if something breaks
One structure I’m considering:
architecture/
README.md
principles.md
tasks.md
inventory/
current-state.md
capabilities.md
permissions-and-boundaries.md
integrations.md
file-structure.md
changelog.md
open-questions.md
skills/
map-openclaw-setup.md
Idea:
- keep it as a pilot first
- don’t change core files or OpenClaw config yet
- use
map-openclaw-setup.mdas an.mdplaybook for inspecting/updating the inventory - update
inventory/changelog.mdwhen setup changes
Questions:
- Keep this inside a pilot folder first, or put inventory at workspace root?
- Should
map-openclaw-setup.mdlive in existingskills/, or inside the pilot until proven useful? - What should a useful setup inventory track so it doesn’t become a markdown junk drawer?
- Any OpenClaw-specific gotchas around project context, skill activation, memory, context bloat, permissions, or rollback?
I’m trying not to over-engineer it, but I want a clean way to track the setup as it evolves. New to openclaw so help appreciated!