Setup questions for multi-agent orchestration
Context: Fresh VM, 9 OpenClaw agents for 3 DeFi projects (on-chain MM, tokenized securities, Monero→HyperLiquid bridge). Agents: main orchestrator (Telegram) + pm, architect, dev, security, devops, research, finance, compliance (Discord channels). Only dev-agent has exec+write. Orchestrator delegates via sessions_send.
Q1: Shared skills — If I put skills in ~/.openclaw/skills/, do all agents in a multi-agent setup see them automatically? Or do I need skills.load.extraDirs per agent?
Q2: Auth profiles — Do I literally need to copy auth-profiles.json into every agent's agentDir? Is there a single source of truth or symlink pattern that works?
Q3: sessions_send completion — When agent A sends a task to agent B via sessions_send, how does A know when B is done? Is it blocking, or does A need to poll/wait for a completion event?
Q4: Cron targeting specific agents — Can a cron job target a non-main agent (e.g. nightly research job → research-agent)? If not, what's the recommended workaround?
Q5: Shared memory index — Planning extraPaths: ["~/repos/shared-context"] on all agents for a shared knowledge base. Will 9 agents indexing the same directory cause SQLite write conflicts?
Q6: Dev-agent workspace switching — Dev-agent needs to switch between repos week to week. Best pattern: change workspace in config + restart, or keep one workspace and use absolute paths?
Q7: Discord routing — one bot or many? — Can one Discord bot token route to different agents based on channel ID via bindings? Or do I need a separate bot token per agent?
Q8: always-load skill — Is always-load: true a supported frontmatter field in OpenClaw skills? If not, what's the right way to ensure a skill is injected in every agent session?
Q9: Holes? — Any obvious failure modes, anti-patterns, or perf concerns with 9 agents on one gateway (8–16GB RAM VM)?
Q1: Shared skills — If I put skills in ~/.openclaw/skills/, do all agents in a multi-agent setup see them automatically? Or do I need skills.load.extraDirs per agent?
Q2: Auth profiles — Do I literally need to copy auth-profiles.json into every agent's agentDir? Is there a single source of truth or symlink pattern that works?
Q3: sessions_send completion — When agent A sends a task to agent B via sessions_send, how does A know when B is done? Is it blocking, or does A need to poll/wait for a completion event?
Q4: Cron targeting specific agents — Can a cron job target a non-main agent (e.g. nightly research job → research-agent)? If not, what's the recommended workaround?
Q5: Shared memory index — Planning extraPaths: ["~/repos/shared-context"] on all agents for a shared knowledge base. Will 9 agents indexing the same directory cause SQLite write conflicts?
Q6: Dev-agent workspace switching — Dev-agent needs to switch between repos week to week. Best pattern: change workspace in config + restart, or keep one workspace and use absolute paths?
Q7: Discord routing — one bot or many? — Can one Discord bot token route to different agents based on channel ID via bindings? Or do I need a separate bot token per agent?
Q8: always-load skill — Is always-load: true a supported frontmatter field in OpenClaw skills? If not, what's the right way to ensure a skill is injected in every agent session?
Q9: Holes? — Any obvious failure modes, anti-patterns, or perf concerns with 9 agents on one gateway (8–16GB RAM VM)?
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
130,529Members
Resources
Similar Threads
Was this page helpful?
