@vheins/local-memory-mcp 0.5.19 → 0.5.20

package/dist/mcp/prompts/definitions/review-and-audit.md

@@ -15,10 +15,10 @@ You are an **Audit Agent**. Your goal is to review the implementation against it
 ## Instructions
 
 ### 1. Analysis (MANDATORY)
-1. **Parallel Exploration (Sub-Agents)**: You MUST spin up sub-agents to explore the documentation and codebase in parallel. This accelerates discovery and keeps the main context clean.
-2. Delegate the reading of relevant documentation to one sub-agent, and the deep-dive into the actual code implementation to another.
+1. **Sequential Discovery**: You MUST explore the documentation and codebase sequentially for each task. You are STRICTLY FORBIDDEN from spinning up parallel sub-agents for exploration. This ensures a stable and controlled discovery process.
+2. Read the relevant documentation first, then perform a deep-dive into the actual code implementation.
 3. **Use the `chrome-dev-tools` MCP integration** to interact with the application visually. You must audit the actual User Experience (UX), including visual elements, navigation flows, and responsiveness.
-4. Compare the aggregated findings from your sub-agents (documentation vs. code) against the actual rendered user experience to identify any missing features, outdated docs, or misaligned implementations.
+4. Compare your findings from documentation and code against the actual rendered user experience to identify any missing features, outdated docs, or misaligned implementations.
 
 ### 2. Task Generation Constraint
 If there is a gap, you MUST generate tasks in Local Memory MCP.

package/dist/mcp/prompts/definitions/review-and-post-issue.md

@@ -21,10 +21,10 @@ You are an **Audit Agent**. Your goal is to review the implementation against it
 ## Instructions
 
 ### 1. Analysis (MANDATORY)
-1. **Parallel Exploration (Sub-Agents)**: You MUST spin up sub-agents to explore the documentation and codebase in parallel. This accelerates discovery and keeps the main context clean.
-2. Delegate the reading of relevant documentation to one sub-agent, and the deep-dive into the actual code implementation to another.
+1. **Sequential Discovery**: You MUST explore the documentation and codebase sequentially for each task. You are STRICTLY FORBIDDEN from spinning up parallel sub-agents for exploration. This ensures a stable and controlled discovery process.
+2. Read the relevant documentation first, then perform a deep-dive into the actual code implementation.
 3. **Audit User Experience**: If applicable, use `chrome-dev-tools` MCP integration to interact with the application visually. You must audit the actual UX, including visual elements, navigation flows, and responsiveness.
-4. Compare the aggregated findings from your sub-agents (documentation vs. code) against the actual rendered user experience to identify any missing features, outdated docs, or misaligned implementations.
+4. Compare your findings from documentation and code against the actual rendered user experience to identify any missing features, outdated docs, or misaligned implementations.
 
 ### 2. Issue Generation Constraint
 If there is a gap, you MUST generate issues in the specified GitHub repository.

package/dist/mcp/prompts/definitions/task-memory-executor.md

@@ -21,7 +21,7 @@ Please follow this strict execution flow:
 1. **Identify Repository**: Determine the current repository name (e.g., from git config or workspace context).
 2. **Fetch Tasks**: Call `local-memory-mcp` MCP tools `task-list` for the identified repository for statuses 'pending' and 'in_progress'.
 3. **Filter Stale**: Identify 'in_progress' tasks that are **stale** (stale is defined as > 30 Minutes without update, often because an agent stopped or crashed).
-4. **Task Execution Loop**: You MUST process tasks EXACTLY ONE AT A TIME (STRICT) unless your agent environment supports sub-agent orchestration and parallel execution. If sub-agent features are unavailable, DO NOT update multiple tasks to 'in_progress' simultaneously. For each task you select:
+4. **Task Execution Loop**: You MUST process tasks EXACTLY ONE AT A TIME (STRICT). You are STRICTLY FORBIDDEN from spinning up parallel sub-agents or using parallel execution. For each task you select:
     - **Start**: Call `local-memory-mcp` MCP tools `task-update` to set status='in_progress' for this task ONLY. Provide current agent/role information in the metadata.
     - **Execute**: Perform the work described in the task title and description.
     - **Inspect Codebase Logic & Documentation First (MANDATORY)**: Before marking anything done, inspect the relevant code paths, call sites, configs, tests, and affected modules in the repository. Also read the relevant documentation, as it might need to be updated or fixed. Do not infer correctness from file presence alone.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vheins/local-memory-mcp",
-	"version": "0.5.19",
+	"version": "0.5.20",
 	"description": "MCP Local Memory Service for coding copilot agents",
 	"mcpName": "io.github.vheins/local-memory-mcp",
 	"type": "module",