opencode-snippets 2.0.1 → 2.1.0

package/skill/snippets/SKILL.md

@@ -10,12 +10,16 @@ Reusable text blocks expanded via `#hashtag` in messages.
 ## Locations
 
 ### Snippets
-- **Global**: `~/.config/opencode/snippet/*.md`
-- **Project**: `.opencode/snippet/*.md` (overrides global)
+- **Global directories**: `~/.config/opencode/snippet/` and `~/.config/opencode/snippets/`
+- **Project directories**: `.opencode/snippet/` and `.opencode/snippets/` (project overrides global, `snippet/` wins over `snippets/`)
 
 ### Configuration
-- **Global**: `~/.config/opencode/snippet/config.jsonc`
-- **Project**: `.opencode/snippet/config.jsonc` (merges with global, project takes priority)
+- **Global file**: `~/.config/opencode/snippet/config.jsonc`
+- **Project file**: `.opencode/snippet/config.jsonc` (merges with global, project takes priority)
+
+IMPORTANT: Snippets live only in those four snippet directories. Check those exact locations. Do not glob anywhere else in the repo or workspace.
+
+IMPORTANT: Config files stay under `snippet/config.jsonc`. The plural `snippets/` support is for snippet markdown files only.
 
 IMPORTANT: When modifying snippet configuration:
 1. Check BOTH locations for existing config files
@@ -35,7 +39,7 @@ Full config example with all options:
 ```jsonc
 {
   // JSON Schema for editor autocompletion
-  "$schema": "https://raw.githubusercontent.com/JosXa/opencode-snippets/v2.0.1/schema/config.schema.json",
+  "$schema": "https://raw.githubusercontent.com/JosXa/opencode-snippets/v2.1.0/schema/config.schema.json",
 
   // Logging settings
   "logging": {
@@ -52,7 +56,10 @@ Full config example with all options:
     "injectBlocks": false,
     // Enable skill rendering with <skill>name</skill> syntax
     // Default: false
-    "skillRendering": false
+    "skillRendering": false,
+    // Enable #skill(name) syntax
+    // Default: false
+    "skillLoading": false
   },
 
   // Hide shell command in output, showing only the result
@@ -152,6 +159,31 @@ Enable in config:
 
 Skills are loaded from OpenCode's standard directories (`~/.config/opencode/skill/` and `.opencode/skill/`).
 
+### Skill Loading (Experimental)
+
+Load OpenCode skills with command-style syntax while keeping the visible message compact:
+
+```md
+Use caveman mode. #skill(caveman)
+<!-- or -->
+#skill("opencode-config")
+```
+
+Enable in config:
+```jsonc
+{
+  "experimental": {
+    "skillLoading": true
+  }
+}
+```
+
+Visible transcript text becomes `↳ Loaded name`. The model receives the full OpenCode-style `<skill_content>` wrapper above that message.
+
+Treat `#skill(...)` as hidden context injection, not inline expansion. User usually sees compact placeholder text, while the model receives an extra injected user message containing the full `<skill_content>` payload. Do not assume one visible bubble can hold different hidden text. The implementation achieves this by inserting an additional hidden message before the visible one.
+
+`#skill(...)` also works when produced by snippet expansion, not only when the user types it directly.
+
 ## Commands
 
 - `/snippet add <name> [content]` - create global snippet