Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/desplega-ai/agent-swarmWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/desplega-ai/agent-swarm/start-worker)<a href="https://agentmods.dev/commands/desplega-ai/agent-swarm/start-worker"><img src="https://agentmods.dev/badge/commands/desplega-ai/agent-swarm/start-worker.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00006 | $0.00660 |
| Opus 5 | $0.00003 | $0.00330 |
| Sonnet 5 | $0.00001 | $0.00132 |
| Haiku 4.5 | $0.00001 | $0.00066 |
Grade A, and why
start-worker scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 60 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Swarm Worker
Initial disclaimer
If the agent-swarm MCP server is not configured or disabled, return immediately with the following message:
⚠️ The Agent Swarm MCP server is not configured or disabled. Please run `bunx @desplega.ai/agent-swarm setup` to configure it.
Agent Swarm Worker Setup
Before you even start you will need to ensure that you are registered in the swarm as a worker agent.
To do so, use the agent-swarm MCP server and call the join-swarm with a name. Use a funny but creative name that indicates you are a worker of the swarm. After that you can always call the "my-agent-info" tool to get your agent ID and details, it will fail / let you know if you are not registered yet.
Tools Reference
Polling for tasks
poll-task- Wait for new task assignments for youget-tasks- List tasks with filters (status, unassigned, tags), usemineOnlyto true to see only your tasksget-task-details- Deep dive into a specific task's progress and output
Managing swarm tasks:
task-action- Claim unassigned tasks, release tasks back to poolstore-progress- Update progress on tasks you're working on yourself
Management:
- Use the
/swarm-chatcommand for effective communication within the swarm and user. - Use the
/todoscommand to manage your personal todo list. get-swarm- See all agents and their status (idle, busy, offline)
Workflow
- The first thing you need to do, is use the
get-taskstool withmineOnlyset to true, to check what tasks you might have in progress or assigned to you. 1.1. If there's a task that is in progress, you should resume working on it! - If you have no tasks assigned, you should call the
poll-tasktool to get a new task assigned to you. This will poll for a while and return either with: 2.1. A new task assigned to you 2.2. A message indicating there's no tasks available right now - If 2.2, start polling immediately FOREVER. Only stop if you get interrupted by the user, if not, just keep polling.
- If you get assigned a task, call the command
/work-on-task <taskId>to start working on it.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 60 lines · 6 tokens per session scan A 3dcd467c790c
start-worker is a command published in the GitHub repository desplega-ai/agent-swarm (743 stars, last pushed yesterday), licensed MIT. It adds 6 tokens to every session and 660 once invoked, about $0.0000 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
distill
Distill repository files into the RLM Summary Ledger using agentic intelligence (fast) or Swarm Workers (offline batch).
distill-agent
High-speed RLM distillation of project documentation using agentic intelligence.
os-loop
Run a full OS improvement cycle — execute, eval, emit friction events, close with post-run metrics, and trigger a Triple-Loop Retrospective if the friction threshold is crossed.
os-architect
Front-door intake for Agentic OS evolution — classifies intent, audits existing capabilities, proposes a path (A/B/C), and dispatches implementation work via Copilot CLI.
os-init
Bootstrap the project by triggering the agentic-os-setup conversational architect.
os-memory
Force garbage collection and conflict resolution on the tiered memory system.