# Connecting private `hermes-soul` to Layered Soul

`hermes-soul` and `layered-soul` have different jobs.

| Repo           | Visibility                          | Purpose                                                                                                     |
| -------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| `hermes-soul`  | private                             | Hermes-native runtime backup from Debby Linux: sessions, sanitized config, scripts, cron, memories, skills. |
| `layered-soul` | public/template or curated instance | Harness-agnostic identity and reviewed context with no secrets or raw sessions.                             |

Do not mirror `hermes-soul` into `layered-soul`. Use a review/export workflow.

## Recommended workflow

1. Back up Hermes normally into the private `hermes-soul` repo.
2. Run the planner from this repo:

   ```sh
   python3 scripts/layered_soul.py plan-private-source \
     examples/private-sources/hermes-soul.example.json \
     --source-root ~/hermes-soul
   ```

3. Review the candidate memory and skill files locally.
4. Write sanitized summaries into this repo:

   ```text
   USER.md
   memories/curated/hermes-memory-summary.md
   skills/<selected-skill>/SKILL.md
   ```

5. Validate before committing:

   ```sh
   python3 scripts/layered_soul.py validate .
   ```

## What may move across

- Reviewed user context from `memories/USER.md`.
- Reviewed memory summaries from `memories/MEMORY.md`.
- Selected `skills/**/*.md` that are safe and useful across harnesses.
- Converted cron/task ideas after they are rewritten as explicit Ops manifests.

## What must stay private

- Raw `sessions/` chat logs.
- `config.yaml` and platform runtime config.
- `channel_directory.json`.
- Tokens, API keys, auth files, browser profiles, lock files, caches, bytecode.

## Example source config

See:

```text
examples/private-sources/hermes-soul.example.json
```

The config is safe to publish because it describes the boundary, not the private
content.