- Plugin manifest (plugin.yaml) with pre_gateway_dispatch + on_session_start hooks - /identity slash command for config management - Honcho injector patch docs (6 lines in plugins/memory/honcho/__init__.py) - Config file at /opt/data/identity-config.json for persistence - Kanban task body convention: context_peer: <name> - Sample config with thierry/catherine mappings Architecture: user-installed plugin (hooks + CLI) + 6-line Honcho bundled plugin change
36 lines
1.1 KiB
Markdown
36 lines
1.1 KiB
Markdown
# Enforcing context_peer in the kanban-worker skill
|
|
|
|
Profiles (Claire, Ashley, Finn, Matt) must include `context_peer: <name>` in
|
|
every `kanban_create` call. This document covers how to enforce it.
|
|
|
|
## Option 1: Profile system prompt (recommended, no code change)
|
|
|
|
Add to each profile's system prompt or skill config:
|
|
|
|
> When creating a kanban task with `kanban_create`, you MUST include
|
|
> a `context_peer: <name>` metadata block in the task body. The peer
|
|
> must be the intended user's Honcho peer name (e.g., `thierry`,
|
|
> `catherine`), not your own profile name.
|
|
>
|
|
> Format:
|
|
> ```metadata
|
|
> context_peer: thierry
|
|
> ```
|
|
|
|
## Option 2: kanban-create wrapper tool
|
|
|
|
Create a custom tool in `/opt/data/hermes-tools/` that wraps `kanban_create`
|
|
and rejects calls without `context_peer` in the body.
|
|
|
|
This approach requires the persistent tools volume (already used for QET,
|
|
Gitea, Ollama tools) and a tool registration in the Docker entrypoint.
|
|
|
|
## Option 3: Slash command validation
|
|
|
|
The `/identity` command includes a `validate` subcommand that checks recent
|
|
kanban tasks for missing `context_peer` fields.
|
|
|
|
```bash
|
|
/identity validate
|
|
```
|