- Docs
- Hermes
- Best Practices
- Agent Capability Audit - the 4 must-install plugins test
Agent Capability Audit - the "4 must-install plugins" test¶
Last updated: 2026-10-30
A popular Hermes-ecosystem creator published a checklist video: "Do not run Hermes Agent another week without installing these 4 plugins first." The video names no products - it names four capabilities. We transcribed the video and audited the CorpusIQ stack against each one.
The 4 capabilities (from the video)¶
- Live cost control - monitors what the agent actually spends on model usage while it runs, so you find out before the bill arrives.
- Model picker - switch models on the fly instead of being locked into the setup-time configuration: cheap models for grunt work, expensive ones when the task demands it.
- Goal management - the agent holds objectives across sessions instead of thinking one conversation at a time.
- Agent bridge - lets your agent hand work to your other agents instead of you copying context around manually. The same suite includes WhatsApp and Discord voice bridges.
CorpusIQ stack mapping¶
| Capability | What we run | Verdict |
|---|---|---|
| Live cost control | Token-usage monitoring on every session plus cost attribution tracking across the fleet | Covered |
| Model picker | A model router with a cost ladder (cheap primary, escalation tiers for complex work) | Covered |
| Goal management | Semantic memory + session handoff + knowledge-graph notes so objectives persist across sessions | Covered |
| Agent bridge | MCP server for tool sharing across agents + email delivery for human-facing reports | Partial - no WhatsApp voice bridge (Hermes covers Discord voice channels natively) |
Takeaway¶
The checklist is satisfied by Hermes-native capabilities (cost tracking, model routing, persistent memory, MCP) - no third-party plugin is required to cover the four categories. The one genuine gap versus the creator's suite is the WhatsApp voice bridge: Hermes voice mode covers the CLI and live conversations in Discord voice channels natively, so Discord is not a gap. The CorpusIQ fleet delivers its agent-to-human output by email. If agent reach on WhatsApp becomes a priority, that bridge is the addition to make.