Agent Ecosystem¶
RemoteJuggler includes a fully operational agent plane that autonomously maintains, monitors, and secures the repository. Three AI agents -- OpenClaw, HexStrike, and Claude Code -- run campaigns through a shared MCP gateway, with results published publicly to GitHub Discussions.
Architecture¶
graph TB
subgraph "Civo K8s (fuzzy-dev)"
GW[rj-gateway<br/>43 MCP tools]
OC[OpenClaw Agent<br/>Claude Sonnet]
HS[HexStrike Agent<br/>Claude Opus]
CR[Campaign Runner<br/>Sidecar]
SE[Setec<br/>Secret Store]
AP[Aperture<br/>LLM Proxy]
end
subgraph "GitHub"
GH[tinyland-inc/remote-juggler]
DI[Discussions]
IS[Issues]
PR[Pull Requests]
end
CR -->|dispatch| OC
CR -->|dispatch| HS
OC -->|MCP tools| GW
HS -->|MCP tools| GW
GW -->|secrets| SE
OC -->|LLM calls| AP
HS -->|LLM calls| AP
CR -->|publish results| DI
CR -->|create findings| IS
CR -->|agent PRs| PR
AP -->|usage metrics| GW
Agents¶
| Agent | Model | Role | Replicas |
|---|---|---|---|
| OpenClaw | Claude Sonnet | General-purpose analysis, dependency audits, issue triage | 1 |
| HexStrike | Claude Opus | Security scanning, CVE monitoring, credential exposure | 0 (on-demand) |
| Claude Code | Direct MCP | Gateway health checks, MCP regression testing | N/A (tools only) |
Campaign Runner¶
The campaign runner is a Go sidecar in the OpenClaw pod that orchestrates all campaign execution:
- Trigger evaluation: Cron schedules, webhook events, manual triggers, dependency chains
- Agent dispatch: Routes campaigns to the appropriate agent (OpenClaw, HexStrike, or direct MCP)
- Result collection: Stores results in Setec with timestamped history
- Feedback: Creates/closes GitHub issues based on findings
- Publishing: Posts sanitized results to GitHub Discussions
- Kill switch: Global emergency halt via Setec key
Live Status¶
Agent Ecosystem Status¶
Last updated: 2026-02-27 04:00 UTC
| Campaign | Agent | Last Run | Status | Key Metric |
|---|---|---|---|---|
| Gateway Health | gateway-direct | <1h ago | PASS | -- |
| Dependency Audit | openclaw | -- | -- | -- |
| Credential Scan | hexstrike | -- | -- | -- |
| Gateway Smoketest | openclaw | -- | -- | -- |
| MCP Regression | gateway-direct | -- | -- | -- |
| Audit Completeness | cross-agent | -- | -- | -- |
View all reports in Discussions
Security Model¶
- No secrets in public output: All Discussion posts are sanitized (regex + entropy + URL patterns)
- Budget enforcement: Every campaign has
aiApiBudget.maxTokens - Read-only default: All campaigns are
readOnly: trueunless explicitly opted out - Kill switch:
remotejuggler/campaigns/global-killin Setec halts all campaigns - Approval gates: Security findings create issues; humans decide action