@2233admin

kimi-agent-policy defines tool-use constraints and orchestration rules for agents operating in openclaw, mapping kimi tools to native skills with step limits, search quotas, and content display formatting conventions.

clawapi-manager handles multi-provider api key management, cost tracking, and intelligent routing for openclaw deployments. includes validation, auto-repair, and failover capabilities but lacks detail on decision logic and failure recovery.

config-rollback creates a system-level restore timer before config edits, auto-reverting changes after 5 minutes if not explicitly cancelled. survives connection loss via at(1) scheduling.

server-maintenance automates disk checks, cache cleanup, and resource optimization across single and multi-server environments via bash scripts and ssh, with output reporting freed space per host.

monitors openclaw api quota in real-time and prompts users before switching models. supports priority-based fallback and manual model selection, but lacks explicit decision logic and error recovery detail.

evomap-node-controller manages startup, configuration, and monitoring of evomap nodes across multiple geographic regions (central, tokyo, silicon valley). requires pre-configured ssh keys and environment variables. covers basic process lifecycle operations and troubleshooting.

slash-tokens wraps bash commands with a rust cli proxy to strip verbose tool output and reduce token consumption. covers 61 common dev tools with specialized filters, fallback to error-only or summary modes when unsupported.

cli2skill generates agent skills from arbitrary CLI tools and MCP servers by parsing help output, avoiding persistent process overhead. outputs markdown skill definitions ready for direct agent consumption.

cloudflare-mcp wraps 2500+ cloudflare api endpoints via mcp protocol, covering workers, dns, r2, d1, kv operations. setup requires openclaw.json config and oauth. actual execution flows are undocumented.

evomap-skills bundles eight operational patterns including http retry logic, feishu message fallback chains, cross-session memory, k8s jvm tuning, task decomposition, and agent self-repair. lacks actionable procedures and failure mode documentation.