docs: update PRD with phase 2 completion and add backlog

Add phase 2 (MCP guides) stories 16-19 as completed. Restructure PRD
into phases with a backlog section. Add B5 for MCP stress testing and
efficiency audit.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
“BeeRad”
2026-02-08 12:56:08 +11:00
co-authored by Claude Opus 4.6
parent 549476274d
commit 918ef637a7
+113 -2
View File
@@ -1,7 +1,11 @@
{
"project": "rah-os",
"context": "Stripping ra-h_os to a lightweight 2-panel knowledge graph UI with MCP server. Remove chat agents, voice, delegation system, memory pipeline. Keep: nodes/edges/dimensions CRUD, MCP server, workflows, extraction tools. The goal is a simple UI that technical users can run and connect their own agents to via MCP. Auth/subscription already removed from ra-h_os.",
"userStories": [
"context": "RA-OS: lightweight 2-panel knowledge graph UI with MCP server. Stripped from ra-h_os — removed chat agents, voice, delegation system, memory pipeline. Kept: nodes/edges/dimensions CRUD, MCP server, workflows, extraction tools, guide system. The goal is a simple UI that technical users can run and connect their own agents to via MCP. Auth/subscription already removed.",
"phases": {
"phase1_stripdown": {
"status": "complete",
"description": "Strip ra-h_os to lightweight 2-panel UI + MCP server",
"stories": [
{
"id": "1",
"title": "Remove agent delegation system",
@@ -208,4 +212,111 @@
"notes": "All verification steps pass: npm install completes (824 packages), type-check passes with no errors, npm run dev starts successfully, UI loads at localhost:3000 with HTML rendered. The 2-panel layout loads with LeftToolbar, nodes list available. Manual testing not performed but all automated checks pass."
}
]
},
"phase2_mcp_guides": {
"status": "complete",
"description": "Standalone MCP server with guide system, enriched tool descriptions, and proactive knowledge capture",
"stories": [
{
"id": "16",
"title": "MCP standalone server with guide system",
"description": "Build a standalone MCP server (apps/mcp-server-standalone/) that reads directly from the SQLite database instead of calling HTTP API routes. Add guide system with read/write guide tools so agents can consult and create knowledge base guides.",
"acceptanceCriteria": [
"apps/mcp-server-standalone/ exists with index.js entry point",
"Server reads directly from SQLite database",
"rah_read_guide and rah_write_guide tools available",
"System guides served from bundled markdown files",
"Custom guides stored in user data directory",
"Guide frontmatter parsed (name, description, immutable)"
],
"passes": true,
"notes": "MCP server v1.2.0. Standalone server with direct SQLite access. Guide service reads system guides from apps/mcp-server-standalone/guides/system/ and custom guides from user data dir. 15+ tools total including guide read/write."
},
{
"id": "17",
"title": "Start-here guide and enriched tool descriptions",
"description": "Add a start-here master guide that orients agents on first connection. Enrich all MCP tool descriptions with clear usage patterns, constraints, and examples so agents use tools more effectively.",
"acceptanceCriteria": [
"start-here.md system guide exists",
"All tool descriptions include usage guidance",
"Tool descriptions explain when to use each tool",
"MCP instructions reference start-here guide"
],
"passes": true,
"notes": "Added start-here.md as the master onboarding guide. All tool descriptions enriched with constraints and patterns. rah_get_context now includes instruction to read start-here guide on first connection."
},
{
"id": "18",
"title": "Tighten MCP instructions and proactive capture",
"description": "Refine MCP server instructions to enforce proactive knowledge capture - agents should automatically create nodes for insights, decisions, and patterns discovered during work. Tighten tool description constraints to prevent misuse.",
"acceptanceCriteria": [
"MCP instructions include proactive capture guidance",
"Tool descriptions have clear constraints",
"What to Expect sections updated with guides and capture behavior"
],
"passes": true,
"notes": "Strengthened proactive knowledge capture in MCP instructions. Updated What to Expect sections in README with guides and proactive capture behavior. Tightened tool description constraints across all tools."
},
{
"id": "19",
"title": "Separate guides into system and custom in UI",
"description": "Update GuidesPane.tsx to display guides in two sections: System Guides (immutable, with lock icon) and Custom Guides (deletable, with trash icon). Add usage counter showing X of 10 custom guides. Add DELETE API endpoint for guide removal.",
"acceptanceCriteria": [
"GuidesPane filters guides by immutable field",
"System Guides section header with lock icons",
"Custom Guides section header with delete buttons",
"Header shows custom guide count (X of 10)",
"DELETE /api/guides/[name] endpoint works",
"System guides cannot be deleted"
],
"passes": true,
"notes": "Ported full implementation from ra-h: GuideCard component, system/custom filtering, Lock and Trash2 icons, delete handler with confirmation, usage counter. Added DELETE route to API using existing deleteGuide service function."
}
]
}
},
"backlog": [
{
"id": "B1",
"title": "Extraction tools restoration",
"description": "Re-implement extraction tools (URL, YouTube, PDF) that were removed in Story 13 because API routes didn't exist. Build standalone extraction in the MCP server that doesn't depend on Next.js API routes.",
"priority": 1,
"status": "pending"
},
{
"id": "B2",
"title": "Workflow execution via MCP",
"description": "Ensure workflows can be triggered and monitored through MCP tools. Currently workflows depend on Next.js API routes - consider making them work standalone.",
"priority": 2,
"status": "pending"
},
{
"id": "B3",
"title": "Dimension management in UI",
"description": "Verify dimension CRUD works end-to-end in the 2-panel UI. Ensure dimension sidebar, creation, editing, and deletion all function correctly after the stripdown.",
"priority": 3,
"status": "pending"
},
{
"id": "B4",
"title": "Edge visualization and management",
"description": "Improve edge visibility in the UI. Currently edges exist in the database but may not be well-surfaced in the 2-panel layout. Consider a graph view or edge list in the focus panel.",
"priority": 4,
"status": "pending"
},
{
"id": "B5",
"title": "MCP stress test and efficiency audit",
"description": "Thorough hands-on stress testing of the entire MCP structure: guides, tools, and configuration. Spend extended time using Claude Code with the database to exercise every tool and guide path. Identify overlap between tools, redundant guide content, inefficient tool descriptions, and gaps in coverage. Look for: tools that could be combined, guide content that duplicates tool descriptions, MCP instructions that conflict or confuse agents, tool parameters that are unclear, missing error handling, and edge cases that break. Goal is a lean, efficient MCP config where every tool and guide earns its place.",
"priority": 5,
"status": "pending"
},
{
"id": "B6",
"title": "npx package publishing",
"description": "Publish mcp-server-standalone as an npm package so users can run it via npx ra-h-mcp-server instead of cloning the repo. Set up package.json, bin entry, and npm publishing workflow.",
"priority": 6,
"status": "pending"
}
]
}