ari-time-research

name	ari-time-research
description	Research ARI's memory and knowledge over time windows
triggers	["/last30days","/lastweek","/lastmonth","last N days","what did we learn","recent decisions","what happened since"]

Time-Windowed Research Skill

Query ARI's memory and knowledge base over specific time periods. Useful for:

• Reviewing recent decisions and their rationale
• Finding patterns in learned behaviors
• Preparing briefings and reports
• Understanding project evolution

Triggers

Command	Time Window
/lastweek	Past 7 days
/last30days	Past 30 days
/lastmonth	Past 30 days
"last N days"	Past N days
"since [date]"	Since specific date

Usage

Basic Queries

/lastweek patterns

Returns patterns learned in the last 7 days.

/last30days decisions

Returns decisions made in the last 30 days with reasoning.

what did we learn since Monday?

Natural language time query.

Domain Filters

You can filter by domain:

• patterns - Learned patterns and conventions
• decisions - Decisions and their rationale
• fixes - Bug fixes and solutions
• docs - Documentation changes

Example:

/lastweek decisions domain:architecture

Confidence Threshold

Filter by confidence level (0-1):

/last30days patterns minConfidence:0.7

Implementation

This skill uses the ari_memory_time_search MCP tool:

const results = await ari_memory_time_search({
  startDate: '2024-01-01',
  endDate: '2024-01-08',  // Optional, defaults to now
  domain: 'patterns',
  minConfidence: 0.5,
  limit: 50,
});

Workflow

1. Parse time window from command/query
2. Call MCP tool with appropriate parameters
3. Group results by domain or type
4. Synthesize findings into actionable insights
5. Optionally enrich with WebSearch for current best practices

Example Output

## Research: Last 7 Days

### Decisions (3)
1. **ADR-009: Use Zod for all schemas** (Jan 5)
   - Reasoning: Runtime validation, TypeScript integration
   - Confidence: 0.95

2. **Chose Vitest over Jest** (Jan 3)
   - Reasoning: ESM support, faster, TypeScript-native
   - Confidence: 0.90

### Patterns Learned (5)
- EventBus for all cross-layer communication
- Audit all state changes
- Trust levels on every message
...

### Fixes Applied (2)
- Fixed hash chain verification on startup
- Resolved memory leak in websocket handler

Integration with Briefings

Results from time research can feed into:

• Morning briefings (week context)
• Evening summaries (day context)
• Weekly reviews (full week context)

Notes

• High-confidence results (>0.8) are highlighted
• Results are sorted by date (newest first)
• Maximum 50 results per query (paginate for more)