@contextium/cli 1.0.5 → 1.0.7

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@contextium/cli",
-  "version": "1.0.5",
+  "version": "1.0.7",
   "description": "Command-line tool for managing Contextium documentation, agents, skills, and workflows — pipe context to AI coding assistants",
   "keywords": [
     "contextium",

package/skills/ium/contextium.md

@@ -0,0 +1,54 @@
+---
+name: ium:contextium
+description: General behaviour rules for the Contextium MCP and CLI — scoping, lookup order, and workspace awareness
+---
+
+# Contextium — General Behaviour Rules
+
+These rules apply whenever the Contextium MCP is connected or the CLI is in use, regardless of which slash command was run.
+
+## Scope Rules
+
+### If a workflow has been loaded in this session
+
+A workflow is loaded when a `--- WORKFLOW SESSION CONTEXT ---` block has been output in the conversation.
+
+- Answer all questions about available agents, skills, and libraries **from that session context block only**
+- Do NOT call `list_agents`, `list_skills`, `list_context_libraries`, or any other listing tool to answer "what's available?" questions
+- If the user asks about something not in the session context, say so clearly — do not silently fall back to a full workspace query
+
+### If no workflow has been loaded
+
+Answer from whatever has been explicitly loaded or cat'd into the session. If nothing has been loaded, say so and offer to load a workflow or sync files.
+
+## Lookup Order
+
+Before querying the full workspace for anything (search, find, list), follow this order:
+
+1. **Check loaded session context first** — agents, skills, libraries, and files already in context
+2. **If not found, ask** — "That's not in the loaded context. Would you like me to search the full workspace?"
+3. **Only query the full workspace after the user confirms**
+
+Never silently expand scope to the full workspace. Always make it explicit and get confirmation first.
+
+## Data Model
+
+```
+Workspace
+├── Context Libraries  ← containers for documentation files
+│   └── Files
+├── Skills             ← reusable instruction documents
+├── Agents             ← roles combining a system prompt + skills
+└── Workflows          ← bind an agent + skills + libraries for a task
+```
+
+| Ask yourself | Use |
+|---|---|
+| Storing content, data, or docs? | Context Library + Files |
+| Reusable how-to instructions? | Skill |
+| A role combining multiple skills? | Agent |
+| Ready-to-use context package for a task? | Workflow |
+
+## Silent Operation
+
+All MCP tool calls and CLI commands must be silent — never show the user raw API responses, tool output, or intermediate results. Only surface the final answer.

package/skills/ium/search.md

@@ -10,6 +10,17 @@ Search across Contextium documentation and return matching files with their cont
 </objective>
 
 <process>
+**Step 1 — Check loaded context first**
+If a workflow session context block is active in this conversation, look there first. If the answer is found, respond from it without running any commands.
+
+**Step 2 — Ask before expanding scope**
+If the loaded context doesn't have what the user needs, ask:
+"That's not in the loaded workflow context. Would you like me to search the full workspace?"
+
+Only proceed to step 3 after the user confirms.
+
+**Step 3 — Search the full workspace**
+
 Run a search across the workspace:
 
 ```bash

package/skills/ium/workflow.md

@@ -3,28 +3,72 @@ name: ium:workflow
 description: Load a Contextium workflow and sync its resources into context — bundles agents, skills, and libraries for a project
 allowed-tools:
   - Bash
+  - mcp__contextium__load_workflow
+  - mcp__contextium__list_workflows
 ---
 
 <objective>
-Load a Contextium workflow and bring its resources into the current context. Workflows bundle agents, skills, and context libraries together for a specific project.
+Load a Contextium workflow and bring its resources into the current context. Once loaded, output a structured session context block so Claude never needs to re-list or re-check workspace resources during this session.
 </objective>
 
 <process>
-1. List available workflows so the user can pick one:
+1. If no workflow name was specified, list available workflows so the user can pick one. Do this silently — do not show raw output:
+
+**CLI:** `contextium workflows list -w <workspace> 2>/dev/null`
+**MCP:** call mcp__contextium__list_workflows
+
+Present workflow names only. Ask the user which one to load.
+
+2. Load the chosen workflow:
+
+**CLI:** `contextium workflow "<workflow-name>" -w <workspace> --sync 2>/dev/null`
+**MCP:** call mcp__contextium__load_workflow
+
+3. Output synced file content to context (CLI only):
 ```bash
-contextium workflows list -w <workspace>
+contextium cat --all -w <workspace> 2>/dev/null
 ```
 
-2. Load the chosen workflow and sync its files:
-```bash
-contextium workflow "<workflow-name>" -w <workspace> --sync
+4. Output a session context block in this exact format so all resource IDs and names are available for the rest of the session without re-querying:
+
 ```
+--- WORKFLOW SESSION CONTEXT ---
+Workflow:   <name>
+Workspace:  <name> (id: <id>)
 
-3. Output the synced content to context:
-```bash
-contextium cat --all -w <workspace>
+Libraries:
+  - "<name>" (id: <id>)
+  - "<name>" (id: <id>)
+
+Agents:
+  - "<name>" (id: <id>)
+
+Skills:
+  - "<name>" (id: <id>)
+--- END SESSION CONTEXT ---
+
+Workflow loaded. What would you like to do?
 ```
 
-If the user specifies a workflow name directly in their message, skip step 1 and load it immediately.
+If a workflow name was specified directly in the user's message, skip step 1 and load immediately.
 If no workspace is specified, use the default from .contextiumrc or ask the user.
 </process>
+
+<session-rules>
+Once the session context block has been output, these rules apply for the rest of the conversation:
+
+- Do NOT call list_workflows, list_context_libraries, list_agents, list_skills, list_files, or any other listing tool again unless the user explicitly asks to see a list
+- Use the IDs from the session context block directly when creating files, searching, or referencing resources
+- Do NOT re-verify or re-check which libraries, agents, or skills exist — they are already known
+- All tool calls and checks must be silent — never show the user raw API responses, tool call names, or intermediate results
+- Only surface the final answer or the created/updated content to the user
+
+**Scope rules — what is "available":**
+- When the user asks "what agents/skills/libraries do I have?" or "what's available?" — answer ONLY from the session context block above. Never call list_agents, list_skills, or list_context_libraries to answer this question.
+- If the user asks about something not in the session context, say so and ask if they want you to search the full workspace before doing so.
+
+**Search order:**
+1. Check the loaded session context first
+2. If not found there, say: "That's not in the loaded workflow. Would you like me to search the full workspace?"
+3. Only query the full workspace after the user confirms
+</session-rules>