Phase 1 (MVP): Memory and ticket pattern analysis

Tinker

Home Epics Phase 1 (MVP): Memory and ticket pattern analysisEdit

Cancel

Title *

Project *

Description

**Goal:** Enable the Researcher to browse global data efficiently and track what it has already analyzed to avoid loops.

## Background
This aligns with the "Proactive" vs "Reactive" shift. Using `researcher_reviewed_at` is the correct architectural choice - it decouples "Content Updates" (user changes description) from "Analysis Status" (AI has read it).

## Implementation Plan

### 1. Database Schema Changes
Migration: Add `researcher_reviewed_at` (datetime, indexed) to `tickets` and `memories` tables.

```ruby
add_column :tickets, :researcher_reviewed_at, :datetime
add_index :tickets, :researcher_reviewed_at

add_column :memories, :researcher_reviewed_at, :datetime
add_index :memories, :researcher_reviewed_at
```

### 2. MCP Tool Updates (agent-bridge)

**list_memories (New/Fix):**
- Args: `limit` (int), `sort_by` (created/reviewed), `direction` (asc/desc), `reviewed_before` (timestamp/null)
- Logic: `Memory.where(researcher_reviewed_at: nil).limit(50)` allows catching up on backlog

**list_tickets (Update):**
- Args: Add `reviewed_before` (timestamp) and status filters
- Logic: Allow fetching "Closed tickets that I haven't reviewed yet" to learn from past solutions

**mark_as_reviewed (New Tool):**
- Args: `type` ("ticket" | "memory"), `ids` (List[int])
- Logic: Bulk updates the `researcher_reviewed_at` timestamp to `Time.now`
- Usage: Agent calls this after generating proposals (or deciding no proposal is needed)

### 3. Researcher Loop Logic (Prompt Strategy)
The Researcher's behavior becomes a simple state machine:

1. **Fetch:** `list_tickets(status: 'closed', reviewed_before: null, limit: 50)`
2. **Analyze:** Find patterns (e.g., "5 tickets about Docker")
3. **Act:**
   - Found issue? -> `create_proposal(...)`
   - Nothing found? -> `mark_as_reviewed(ids: [...])`
4. **Repeat:** For memories

## Acceptance Criteria
- [ ] Schema migration applied (`researcher_reviewed_at`)
- [ ] `list_memories` tool functional with "unreviewed" filter
- [ ] `mark_as_reviewed` tool functional
- [ ] Researcher observed successfully "clearing the queue" of unreviewed items in logs