How Context Works
The system’s secret weapon: giving each part of the system exactly the right information at the right time. Not too much, not too little.
The Core Idea
Analogy: Think of a film production. Everyone on set shares the same production bible (the genre, the tone, the schedule). But the director gets the creative vision document, the cinematographer gets the shot list, and the editor gets the pacing notes. Nobody gets flooded with everything — each person gets what they need.
The agent system works the same way. Context flows in three tiers:
┌─────────────────────────────────────────────┐
│ Tier 1: About YOU │
│ Your profile, preferences, goals │
│ Every agent always sees this │
└─────────────────────────────────────────────┘
┌─────────────────────────────────────────────┐
│ Tier 2: About THIS JOB │
│ Shared standards for this workflow │
│ All agents in this workflow see this │
└─────────────────────────────────────────────┘
┌─────────────────────────────────────────────┐
│ Tier 3: About THIS ROLE │
│ Specific instructions for this step │
│ Only this agent sees this │
└─────────────────────────────────────────────┘Tier 1: About You (Global Context)
This is your personal profile — who you are, what you care about, how you work. It loads automatically and every agent always has access to it.
What’s in it:
- Your identity, values, and communication style
- Your goals and priorities
- Your constraints (energy patterns, schedule preferences)
- A map of everything in the system (so agents can find skills and files)
- Your life domains and how they’re organized
Why it matters: This is what makes every agent feel like YOUR assistant. Without it, you get generic responses. With it, everything is tailored to you.
How It Loads
Some parts load automatically (your core rules and preferences). Other parts load on demand when you tell the system what you’re working on:
| What You Say | What Loads |
|---|---|
| Just start a conversation | Core rules and preferences (automatic) |
| “I’m working on coding” | + Your coding projects, tech preferences, conventions |
| ”I’m working on finance” | + Your portfolio, investment thesis, financial goals |
| ”I’m working on health” | + Your health context, fitness goals, energy patterns |
This keeps things focused. When you’re coding, you don’t need your investment thesis loaded. When you’re reviewing finances, you don’t need your coding conventions.
Tier 2: About This Job (Workflow Context)
When a workflow runs, all the agents involved share a common “playbook” — the goals, standards, and rules for this specific job.
What’s in it:
- The workflow’s objective (what “done” looks like)
- Quality standards (what good output looks like)
- Domain knowledge (best practices for this type of work)
- Progress tracking (what’s been done so far, what’s left)
Analogy: Like a project brief that every team member reads before starting. It aligns everyone on what they’re building, what quality bar they’re hitting, and what constraints they’re working within.
Real Example: Weekly Review Workflow
When you run a weekly review, the workflow context tells every agent involved:
- “We’re producing a weekly review covering these 5 areas”
- “Use this format and this level of detail”
- “Compare to last week’s numbers”
- “Flag anything that’s changed more than 20%”
Every agent in the workflow gets this shared context. They’re all working toward the same goal with the same standards.
Tier 3: About This Role (Agent Context)
Each individual agent gets additional context specific to their role in the workflow. This is the most focused and specific layer.
What’s in it:
- This agent’s specific assignment
- Input from previous steps (what to work with)
- Output expectations (what to produce)
- Skills available for this step
Analogy: Like a job brief for a specific crew member. “You’re the researcher. Here’s what the client wants to know. Here are the sources to check. Deliver a summary in bullet points by this afternoon.”
Real Example: Inside the Weekly Review
| Agent | Their Context (Tier 3) |
|---|---|
| Data Collector | ”Pull metrics from calendar, email, and tasks. Output as structured data.” |
| Analyst | ”Here’s the data. Find patterns, compare to last week, highlight anomalies.” |
| Writer | ”Here’s the analysis. Write the review in the user’s voice, use this template.” |
| Quality Checker | ”Here’s the draft. Verify accuracy, completeness, and tone.” |
Each agent sees only what they need. The data collector doesn’t see the writing template. The writer doesn’t see which APIs were queried. This keeps every agent focused and efficient.
Why This Matters
Without tiers: With tiers:
┌──────────────────┐ ┌──────────────────┐
│ Every agent gets │ │ Each agent gets │
│ EVERYTHING │ │ WHAT THEY NEED │
│ │ │ │
│ ──▶ Confusion │ │ ──▶ Focused │
│ ──▶ Slow │ │ ──▶ Fast │
│ ──▶ Poor quality │ │ ──▶ High quality │
└──────────────────┘ └──────────────────┘Without the separation: Every agent in a workflow gets ALL information about EVERY step. A researcher sees formatting instructions meant for the editor. An editor sees raw research prompts. Context gets bloated, agents get confused, quality drops.
With the separation: Each agent gets exactly what it needs. They work faster, stay focused, and produce better results.
How Context Stays Fresh
Context goes stale — your goals change, new skills get added, priorities shift. The system keeps context fresh two ways:
Automated Refresh (Runs Overnight)
Every night, an automated pipeline:
- Collects data from your connected tools (email, calendar, messages)
- Scans your recent conversations for patterns
- Updates your profile with new information
- Syncs everything
You wake up to fresh, accurate context.
Manual Updates (When Things Change)
Some changes can’t wait for the nightly refresh:
- Started a new project? Update your project list.
- Changed a goal? Update your goals file.
- Added a new skill? Update the system map.
Rule of thumb: If the change would affect how agents work with you TODAY, update it now. If it can wait until tomorrow, the automated refresh will handle it.
Learn More
- The 8 Primitives — See how context fits into the full system
- How It Learns — How feedback improves context over time
- Context Architecture (Technical) — File paths, loading mechanisms, and configuration