AetherMind Hermes Plugin

AetherMind adds project-local continuity tools to Hermes. It stores compact, append-only .aem records beside a project so future agent sessions can recover the decisions, corrections, uncertainty, and pressure that matter without preserving full transcripts.

When enabled, the plugin also registers Hermes lifecycle hooks. Fresh Hermes sessions automatically initialize/read the project .aethermind store and inject a short AetherMind continuity block before model calls.

This repository is the Hermes plugin distribution. It is intentionally small:

• plugin.yaml and __init__.py are the Hermes plugin entry point.
• aem_store.py contains the local .aem store behavior.
• skills/aethermind-continuity/ provides the companion Hermes skill.
• docs/ and examples/ describe the public store contract.

Install

hermes plugins install MaverickKB/aethermind-hermes-plugin --enable
hermes plugins list

Restart Hermes after enabling the plugin. In a running Hermes session, /plugins should show aethermind.

After restart, run Hermes from the project directory you want AetherMind to cover. The plugin uses the process working directory as the default project root. You can override it with AETHERMIND_PROJECT_ROOT.

Automatic Use

The plugin registers:

• on_session_start, which creates .aethermind/layers.aem and .aethermind/texture.aem for the current project if they do not exist.
• pre_llm_call, which reads relevant layers and injects an AetherMind continuity block into the current user turn.

That injected block tells the agent to use the aethermind_* tools and to write layers for load-bearing decisions, corrections, discoveries, friction, or uncertainty. The plugin does not fabricate routine task-summary layers; it keeps meaningful layer creation tied to actual project work.

The companion skill is plugin-qualified:

aethermind:aethermind-continuity

Hermes exposes that skill through skill_view. It is not copied into ~/.hermes/skills and does not appear in the flat skills list.

Tools

The plugin registers:

• aethermind_init_store
• aethermind_write_layer
• aethermind_read_layers
• aethermind_write_texture
• aethermind_read_texture
• aethermind_reorient
• aethermind_evaluate_store
• aethermind_export_store
• aethermind_import_layers
• aethermind_integrity_manifest

What AetherMind Is

AetherMind is a continuity substrate, not a general memory database. Use it for small, durable signals that help the next agent understand the project state:

• decisions and constraints that should not be rediscovered;
• corrections to wrong assumptions;
• unresolved uncertainty that changes the next step;
• friction or pressure that shaped the work;
• evidence and verification pointers.

What It Is Not

AetherMind is not a transcript archive, task log, chat history, analytics store, or deployment platform. Keep layers compact and avoid writing secrets, private operator paths, customer data, raw prompts, or high-volume logs.

Validation

From the repository root:

./scripts/ci-local.sh

The validation script compiles the plugin, smoke-tests the Hermes registration surface, verifies lifecycle hook behavior, checks the example store, and scans the public files for common publishing mistakes.

Documentation

• docs/SCOPE.md defines the product boundary.
• docs/AEM_FORMAT.md describes the baseline .aem record format.
• docs/HERMES_PLUGIN.md covers Hermes installation and discovery notes.
• docs/PRIVACY.md describes safe layer hygiene.

License

Apache-2.0. See LICENSE.