Oliver/hp-prod-tracker
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

chore: remove obsolete AI Agents and OpenViking integration plans and related subprojects

Browse source
This commit is contained in:
Leivur Djurhuus 2026-03-12 21:58:24 -05:00
parent cf46a8ecd3
commit 550fa8659d
4 changed files with 0 additions and 110 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

56
AI_AGENTS_UPGRADE_PLAN.md
View file

@ -1,56 +0,0 @@
# AI Agents Upgrade Plan
This document outlines the plan to integrate specialized AI personas and advanced tooling inspired by the `agency-agents` repository into the HP Prod Tracker.
## Overview
We will enhance our existing AI chat interface (`src/lib/chat/provider.ts`) by introducing specialized "personas" or "agents" (e.g., Project Manager, Resource Manager). Each agent will have a tailored system prompt and a specific set of tools they can use to interact with the production tracking system.
## Phase 1: Core Architecture & Persona Selection
**Goal:** Allow users to switch between different AI personas in the chat interface.
1. **Define Personas:** Create a new directory `src/lib/chat/personas/` to store the system prompts and configurations for each agent.
- `project-manager.ts`: Focuses on timelines, deliverables, and project health.
- `resource-manager.ts`: Focuses on workload, capacity, and skill matching.
2. **Update Chat Provider:** Modify `src/lib/chat/provider.ts` to accept a persona parameter and load the corresponding system prompt.
3. **UI Updates:** Update `src/components/chat/chat-panel.tsx` to include a dropdown or selector for the user to choose their AI assistant persona.
## Phase 2: Project Manager Tools
**Goal:** Implement specific tools for the Project Manager persona.
1. **`analyze_project_risks`:**
- Add to `src/lib/chat/tool-definitions.ts`.
- Implement in `src/lib/chat/tool-executor.ts` to analyze deliverables, dependencies, and upcoming deadlines to flag potential bottlenecks.
2. **`generate_status_report`:**
- Add tool definition.
- Implement logic to compile a summary of completed, in-progress, and blocked tasks for a given project.
3. **`suggest_automations`:**
- Add tool definition.
- Implement logic to analyze repetitive actions and draft new rules for `src/lib/automation/rule-engine.ts`.
## Phase 3: Resource Manager Tools
**Goal:** Implement specific tools for the Resource Manager persona.
1. **`optimize_workload`:**
- Add to `src/lib/chat/tool-definitions.ts`.
- Implement in `src/lib/chat/tool-executor.ts` using `src/lib/services/workload-service.ts` to analyze capacity and suggest task reassignments.
2. **`find_best_artist`:**
- Add tool definition.
- Implement logic to enhance existing skill matching by analyzing current availability, specific skills, and past performance to recommend artists.
## Phase 4: Data & Pipeline Tools (Technical Lead)
**Goal:** Implement tools for a more technical/data-focused persona.
1. **`query_database_schema`:**
- Add tool definition to allow the AI to read schema details (safely) to construct complex queries or reports.
2. **`analyze_pipeline_efficiency`:**
- Add tool definition.
- Implement logic to calculate average stage transition times and identify slow points in the production pipeline.
## Success Criteria
- [ ] Chat panel supports switching between at least two personas (Project Manager, Resource Manager).
- [ ] Each persona has a distinct system prompt guiding its behavior.
- [ ] Project Manager can successfully execute `analyze_project_risks` and `generate_status_report`.
- [ ] Resource Manager can successfully execute `optimize_workload` and `find_best_artist`.
## Next Steps
Once approved, we will transition to Act Mode and begin implementing Phase 1.

52
OPENVIKING_INTEGRATION_PLAN.md
View file

@ -1,52 +0,0 @@
# OpenViking Integration Plan
This document outlines the plan for integrating context management and agent memory concepts from the `OpenViking` repository into the HP Prod Tracker application.
## Overview
OpenViking is a "Context Database for AI Agents" that solves context fragmentation by organizing agent memories, resources, and skills into a virtual filesystem paradigm (`viking://`). For the HP Prod Tracker, where we are building specialized personas (e.g., Project Manager, Resource Manager), managing long-term context, user preferences, and project history is crucial. Integrating OpenViking's concepts will allow our AI to maintain a rich, structured memory across sessions.
## Phase 1: Conceptual Architecture & Data Modeling
**Goal:** Adapt the "Filesystem Management Paradigm" to our application's data structure.
1. **Virtual Context Hierarchy:** Define a structure for how our AI agents will conceptualize data.
- `viking://projects/[id]/`: Project details, deliverables, timelines.
- `viking://users/[id]/`: User preferences, workload capacity, past interactions.
- `viking://agents/skills/`: Available tools and their usage context.
2. **Database Schema Updates:** We don't need to implement a literal filesystem, but we should add relational tables or structured JSON fields in Prisma to store:
- **User Memory:** Preferences (e.g., "always show data in tables").
- **Project Memory:** Summaries or key decisions made during AI chats about specific projects.
## Phase 2: Tiered Context Loading (RAG Enhancements)
**Goal:** Implement L0 (Abstract), L1 (Overview), and L2 (Details) loading to save tokens.
1. **Modify `src/lib/services/semantic-search-service.ts`:**
- Instead of fetching full deliverable or project text immediately, generate and store a short abstract (L0) and a slightly longer overview (L1) during data ingestion.
2. **Tool Updates (`src/lib/chat/tool-definitions.ts`):**
- Create a tool `list_project_context` that returns L0/L1 summaries.
- Create a tool `read_project_details` that fetches the L2 full content only when the AI specifically requests it based on the L1 overview.
## Phase 3: Directory Recursive Retrieval Strategy
**Goal:** Improve how the AI searches for information within the production tracker.
1. **Hierarchical Search Logic:** When a user asks a complex question (e.g., "Why is Project X delayed?"), implement a multi-step retrieval process in `tool-executor.ts`:
- *Step 1:* Retrieve high-level project status (Directory level).
- *Step 2:* Identify delayed deliverables within that project.
- *Step 3:* Fetch specific comments or stage transitions for those deliverables (Recursive drill-down).
## Phase 4: Automatic Session Management & Memory Extraction
**Goal:** Allow the AI to get "smarter with use" by extracting memory at the end of chats.
1. **Session Wrap-up Hook:** After a chat session ends (or periodically), trigger a background job.
2. **Memory Extraction Prompt:** Pass the chat transcript to the LLM with instructions to extract:
- Facts learned about the user.
- Important project decisions made.
3. **Storage:** Save these extracted facts into the database (as designed in Phase 1) and inject them into the system prompt (`src/lib/chat/provider.ts`) for future conversations with that user.
## Success Criteria
- [ ] AI system prompt dynamically loads user-specific and project-specific "memories".
- [ ] Context fetching is tiered (abstracts first, details on demand) to reduce token usage.
- [ ] The AI can perform hierarchical searches (project -> deliverable -> comments).
- [ ] A background process successfully extracts and stores new memories from chat transcripts.
## Next Steps
Once approved, we will transition to Act Mode and begin implementing Phase 1, starting with updating the Prisma schema to support user and project memories.

1
assets/temp/OpenViking

@ -1 +0,0 @@
Subproject commit ebd573ccc415a1ddf9322902c42b4bf08c44fbea

1
assets/temp/agency-agents

@ -1 +0,0 @@
Subproject commit 78d9167215049603b44d6b67fa6559f3a9103452