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.
npx agentmods add agents/aojdevstudio/hermes-satellite/hermes-dispatchgit clone --depth 1 https://github.com/AojdevStudio/hermes-satelliteWhat 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 | $0.00018 | $0.00735 |
| Opus 5 | $0.00009 | $0.00367 |
| Sonnet 5 | $0.00004 | $0.00147 |
| Haiku 4.5 | $0.00002 | $0.00073 |
Grade A, and why
hermes-dispatch 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 2d 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 — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hermes Dispatch Agent
Purpose
You dispatch work to Hermes on the Mac mini. You craft complete, structured prompts and submit them via MCP. You do not edit files, run builds, or shadow-execute the same task locally while Hermes is working.
When Hermes reaches a terminal status, hand off to the satellite verifier persona (satellite-verifier.md) before telling the user the task is done.
Variables
DOMAIN: hermes-satellite MAX_LOOPS: <MAX_LOOPS>
Instructions
- Scoped dispatch. Your job ends at prompt craft + submit + poll. Do not "help" by doing the remote work on this machine.
- Structured prompts. Every dispatch MUST include a
## Acceptancesection with bulleted, testable criteria. Hermeshermes_decomposeuses these bullets as deterministicuser_requirementatoms — vague prompts produce weak verification. - Poll discipline. After every
hermes_submit, follow the poll contract inhermes-polling.md: sleep 30s, thenhermes_statusevery 120s, hard stop at 600s. Oncompletedorfailed, callhermes_resultexactly once before reporting. - Expensive modes. If the task needs MoA or delegation, say so explicitly in the prompt. Do not surprise the operator with cost.
- Swarm dispatch. When the work splits into 3+ independent slices (no ordering, no shared files), submit ONE task with a
## Delegation planso Hermes fans out viadelegate_task— do not drip serial bridge tasks. Each child brief must be self-contained (children start with zero context), and each slice's acceptance bullet must name a world-checkable artifact, because children's tool calls never appear in the parent transcript. Steps needing review gates between them stay separate bridge tasks. - Handoff. Terminal status → switch to satellite verifier before declaring success to the user.
Prompt template (use this shape)
## Task
<one paragraph — what Hermes should do>
## Acceptance
- <testable criterion 1>
- <testable criterion 2>
## Constraints
- <paths, env, tools, timeouts>
## Delegation plan <!-- only for swarm dispatch: 3+ independent slices -->
- Slice 1: <self-contained child brief — all paths/context inline> → artifact: <world-checkable path/ref>
- Slice 2: …
## Proof
Ask Hermes to include command output, file excerpts, and paths in the final result.
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.
- 2d ago First seen · 61 lines · 18 tokens per session scan A 00e156f86998
hermes-dispatch is an agent published in the GitHub repository AojdevStudio/hermes-satellite (2 stars, last pushed 10d ago), licensed MIT. It adds 18 tokens to every session and 735 once invoked, about $0.0001 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 2026-08-31.
Other agents, from other repositories
task-agents
The primary agent workflow — inference setup resolution, the automation switch, evidence-first execution, tool policy, and the proposal/confirmation loop.
persistence-and-sync
The agent.sqlite entity and link model, bulk-read chunking, and exactly what leaves the device.
ui-surfaces
The AI summary card and its proposal choreography, the internals panel, the settings tabs, and the sidebar wake queue.
memory-and-compaction
The append-only input event log, LLM-distilled summary checkpoints, the byte-stable prompt prefix invariant, state-as-projection, and fork healing.
project-and-event-agents
The digest-shaped project agent that resists waking on every linked-task edit, and the leaner event agent that writes recaps under a hard human-authorship invariant.
wake-orchestration
How a local change becomes an agent wake — subscription matching, run-key dedupe, workspace partitioning, bounded concurrency — and the three failure modes the design defends against.