Skills System
AstroJOE's SKILL.md-based tool system — procedural knowledge documents that enable real tool execution.
AstroJOE uses SKILL.md files as its tool system. Each skill is a procedural knowledge document that teaches the agent how to execute specific operations via the terminal (curl, python3, jq).
Hermes v0.8.0 ships with 101 built-in skills across 20 categories:
autonomous-ai-agents creative custom data-science devops email gaming general github leisure mcp media mlops note-taking productivity red-teaming research smart-home social-media software-development
How Skills Work
Hermes Agent scans ~/.hermes/skills/ recursively for SKILL.md files. Skills follow a three-level progressive disclosure pattern:
- Level 0 —
skills_list(): Returns name, description, and category (~3k tokens) - Level 1 —
skill_view(name): Loads the full SKILL.md content on demand - Level 2 —
skill_view(name, path): Loads specific reference files individually
Skills are loaded only when needed for the current task, conserving context.
SKILL.md Format
---
name: skill_name
description: One-line description of what this skill does
version: 1.0.0
author: AstroJOE
tags: [search, web, memory]
required_environment_variables:
- SOME_VAR
---
# Skill Name
Description of what this skill does and when to use it.
## When to Use
- Trigger conditions...
## Procedure
### Step 1: ...
```bash
curl -s http://example.com/api ...Pitfalls
- Known issues...
Verification
- How to verify it worked...
## Available Skills
### hedgehog-websearch
Search the web and X/Twitter using Grok 4.20 multi-agent mode with built-in `web_search` and `x_search` tools via the HEDGEHOG gateway.
**Trigger**: User asks to search the web, look up current events, or check a website.
### spacetimedb-memory
Full CRUD on the optx-cortex database — store/recall memories, manage agent state, query gaze attestations and session logs.
**Trigger**: User asks to remember something, recall context, or check agent state.
**Tables**: `memory_entry`, `agent_state`, `gaze_attestation`, `session_log`, `matrix_message`
### aaron-router
Gaze verification sessions, iris analysis, and COG/EMO/ENV tensor classification via the AARON Router on the validator node.
**Trigger**: User asks about gaze verification, JETT Auth, or biometric sessions.
### joe-matrix-relay
Send and receive messages through Matrix/Conduit federation via JOE WebSocket server.
**Trigger**: User asks to send a message to Matrix or relay information.
### agenc-tasks
Orchestrate Solana agent swarms via AgenC — create/claim/complete tasks, policy checks, ZK proofs.
**Trigger**: User mentions tasks, swarm, orchestration, or AgenC.
### hedgehog-memory
Lightweight key-value memory via the HEDGEHOG service (separate from SpacetimeDB).
**Trigger**: Quick memory operations without full SpacetimeDB overhead.
## Self-Maintenance
AstroJOE proactively updates its own configuration as infrastructure changes:
- **SOUL.md** � Updates its identity document when new tools, skills, or gateways are added
- **Skills** � Creates and modifies SKILL.md files to reflect new service endpoints or API changes
- **Dev Logs** � Writes Obsidian-format session logs to the OPTX-Cortex vault after significant changes
This self-maintenance protocol ensures AstroJOE's operational knowledge stays current without manual intervention.
## Creating Custom Skills
Place a new directory with a `SKILL.md` file in the skills mount:skills/ ├── my-new-skill/ │ └── SKILL.md
The skill will be discovered automatically on the next agent restart. Skills execute via the terminal backend (Docker container with curl, python3, jq).