Run a persistent Discord agent fleet without the ops headaches

Where this fits

You have a remote server and want several always-on AI Discord bots—one for knowledge retrieval, one for study material, another for community moderation—each with distinct personas and memories. The discord-agent-fleet skill packages that exact setup: three bots sharing one Python venv, one server, and one codebase, with launchd configuration for automatic restarts. It also documents the traps that will silently kill your bots: power-saving modes that block relaunch, config files that need re-reading, and API usage ceilings.

Why agents benefit

• Multi-persona orchestration: Run three bots on one machine from a single skeleton—each bot reads its own persona file on every response, so personality changes apply without restart.
• Two-layer memory: Short-term conversation is persisted to SQLite across reboots; long-term facts accumulate automatically when the bot writes a MEMORY: line in its reply.
• Channel-aware behavior: Bots answer DMs with full context but only reply in public channels when mentioned, reducing noise and API cost.
• Operational awareness built-in: The skill documents the exact launchd service names and the known failure mode (lowpowermode = 1 prevents process resurrection), so agents can diagnose why a bot went silent.
• Instant persona iteration: Since persona.md is read per response, tweaking tone or rules takes effect immediately—no deploy cycle.

Practical scenarios

A knowledge bot that survives reboot

Your launchd service reports state = not running after a server restart. Instead of guessing, the agent checks for lowpowermode = 1 and the pended nondemand state—the documented pitfall—then advises changing power settings rather than blindly re-running the bot.

Cross-session context for study group conversations

A study bot accumulated 2,000 articles and remembers user-specific context. Between sessions, short-term memory resets while long-term facts persist, so the bot picks up mid-conversation references correctly after a drop.

Controlled community bot deployment

A third bot handles guild and slash-command operations. Because authorization is scoped to specific user and community IDs in the config, you can delegate community-facing behavior without opening the bot to everyone.

Add it to your agent workflow

Install the skill and let your agent scaffold the bot fleet on your remote machine:

npx skills add bam-bam-2/solo-skills --skill discord-agent-fleet

Then point the agent at ssh remote-host, have it create ~/Projects/<project>/, copy the shared skeleton, and register each bot as its own launchd service using the naming convention in the README.

Compare before adopting

Before committing, compare against generic Discord bot templates: check whether a skill documents real operational constraints (power modes, config reload behavior, usage limits) or just shows bot code. Also look for maintenance freshness—does the author keep launchd plists and dependency versions current? And verify workflow fit: does your team actually need multiple personas on one server, or would a single bot with routing suffice? For a broader view of similar agent skills, see the entry for bam-bam-2's discord-agent-fleet.

Why it is worth tracking

This skill stands out for writing down pitfalls most bot tutorials skip—the low-power mode trap alone can save hours of debugging. At 11 stars with a focused scope, it is early-stage but demonstrates the maintainer follows through on documentation quality. Track it if you operate persistent agents on underpowered or headless machines; reevaluate when the repo adds more monitoring or multi-host support, which would signal the maintainer is building beyond a personal setup.