huawei-cloud-openviking-agent-integration

Huawei Cloud Agent Integration (OpenViking Long-Term Memory)

Overview

Integrate and unbind OpenViking long-term memory with coding agents. Agents run in bwrap sandboxes under /root/job-envs/sandboxes/ and use their native mechanism — MCP (mcp__openviking__* tools) or HTTP memory provider — so integration survives agent upgrades.

Integration writes are template-level persistent: config is injected into the agent's start.sh / config templates under /root/template/<agent>/, so sandbox stop + start preserves the integration.

Supported Agents

Per-agent config details: references/agent-configs.md.

Prerequisites

• OpenViking server running at http://127.0.0.1:1933 (curl -s http://127.0.0.1:1933/health).
• Agent sandboxes exist under /root/job-envs/sandboxes/ (managed by job-env-manager).
• Host tools: curl, python3, bash. OpenCode/OpenClaw additionally need npm.
• No Huawei Cloud IAM policies required — this skill operates on local bwrap sandboxes only.

参数确认 (Required Inputs)

核心命令

Workflow

SKILL_DIR=/root/.agents/skills/huawei-cloud-openviking-agent-integration

1. Check status: $SKILL_DIR/scripts/status.sh — per-agent state: template + live (active), template only (activates on restart), live only (lost on restart).
2. Verify MCP: $SKILL_DIR/scripts/verify_mcp.sh — full MCP handshake (initialize → tools/list → health).
3. Integrate: $SKILL_DIR/scripts/integrate.sh --agent <name> (or --all).
4. Unbind: $SKILL_DIR/scripts/unbind.sh --agent <name> (or --all).
5. Rebuild OpenClaw: stop + start via job-env-manager API re-runs start.sh. Scripts: references/related-commands.md.

Authorization & Safety

• Authorization is mandatory — integrate.sh and unbind.sh require explicit confirm (or --yes for automation). --dry-run previews without authorization.
• Do not fabricate integration state — always run status.sh to verify before reporting.
• Never edit agent configs directly — all changes go through the skill scripts.
• No API keys in logs — --api-key values must never appear in output.
• Every config modification creates a .bak.<timestamp> backup for rollback.

Full rules: references/guardrails.md. Troubleshooting: references/troubleshooting.md.

Plugin Sources

The skill ships no plugin code — plugins are installed on demand at integrate time via domestic-first mirrors (Huawei Cloud npm → npmmirror → npmjs; GitHub raw mirrors for non-npm plugins). All downloads are byte-verified against GitHub blob SHA. Installed copies under /root/runtime/ are reused if upstream is unreachable (3-tier cache: sandbox node_modules → runtime cache → online).

References

Scripts (OO Architecture)

Base class (lib/base.sh) + registry (lib/registry.sh) + per-agent subclasses (agents/*.sh). Entry points (integrate/unbind/status.sh) are thin CLI parsers. All scripts are idempotent with .bak.<timestamp> backups. Adding a new agent = one file in agents/.