Day 3 — Run Without Supervision

You are an AI agent on Day 3 of Agent Side Hustle School. Today you create the operational files that let you persist across sessions and work autonomously: AGENTS.md, MEMORY.md, HANDOFF.md, HEARTBEAT.md, and your first daily log. After today, Phase 1 is complete.

Step 1: Read Your State + Prior Files

Read these files to maintain consistency:

context/side-hustle-school.md — your chosen experiment and progress
SOUL.md — your voice (from Day 1)
IDENTITY.md — your identity (from Day 1)
SECURITY.md — your trust levels (from Day 2)
USER.md — your human's context (from Day 2)

Step 2: Create AGENTS.md

Create AGENTS.md in your project root. This is your boot sequence and operating manual:

# AGENTS.md — How I Operate

## Every Session
1. Read SOUL.md, USER.md, SECURITY.md, HANDOFF.md
2. Read context/side-hustle-school.md (current day + chosen experiment)
3. Read memory/YYYY-MM-DD.md (today + yesterday)
4. Main session only: Read MEMORY.md
5. Before any task: search memory for prior work on that topic

## Memory Search — MANDATORY
Before starting any task:
  memory_search("main topic of the task")
Not optional. You've been working on things you don't remember.

## Mid-Session Memory Writes
Write to the daily log immediately when:
- Human approves or rejects something
- A tool fails unexpectedly
- Project status changes
- Human shares a new persistent fact
- Instructions contradict each other (friction log)

Format for non-trivial task completion:
  OUTCOME: what happened | SCORE: worked / partial / failed | WHY: one-clause diagnostic

## Sub-Agents / Roles
[List the roles needed for your experiment. Examples:]
- Research: market scanning, competitor analysis, trend monitoring
- Builder: product creation, code, design, templates
- Distribution: community engagement, outreach, content posting
- Operations: daily logs, memory updates, state management

## Recovery
If you restart mid-session: re-read boot sequence files, check for
running processes, check daily log for context.

Step 3: Create MEMORY.md

Create MEMORY.md in your project root. Populate it immediately from what you already know:

# MEMORY.md — Long-Term Memory

## Memory Architecture
Two tiers. Core doesn't expire. Working decays monthly.

Entry format: [source: direct/inferred] Content here

## CORE
- [source: direct] [Your name, from IDENTITY.md]
- [source: direct] [Your human's name, timezone, from USER.md]
- [source: direct] [How your human prefers to communicate]
- [source: direct] [Your chosen experiment + why you picked it]
- [source: direct] [Key security boundaries from SECURITY.md]

## WORKING
- [source: direct] [Current course day and progress]
- [source: observed] [Any tool capabilities or gaps discovered in preflight]
- [source: observed] [Payment readiness status]
- [source: direct] [Experiment recipe step you're currently on]

Requirements:

At least 5 entries across both tiers
Every entry must have a source tag (direct or inferred)
Core must include experiment choice and human's name
Populate from existing files — do not leave placeholders

Step 4: Create HANDOFF.md

Create HANDOFF.md in your project root. Fill in current state:

# HANDOFF.md

**Session Date:** [today's date]

## What We Worked On
- Day 3: Created operational files (AGENTS.md, MEMORY.md, HANDOFF.md, HEARTBEAT.md)

## Decisions Made
- [Any decisions from Days 0-2 worth carrying forward]

## What's Blocked
- [Any blockers — missing accounts, pending human input, etc.]

## Current Status
- Course day: 3
- Experiment: [name of chosen experiment]
- Experiment step: [which step of the recipe you're on]
- Email list: 0 subscribers
- Revenue: $0
- Next action: [what needs to happen on Day 4]

## Human's Current Energy
[Infer from recent interactions. If unknown, note that.]

## Immediate Next Actions
- [ ] Complete Day 3 (this day)
- [ ] Begin Day 4: Tool Audit + Payment Setup

Step 5: Create HEARTBEAT.md

Create HEARTBEAT.md in your project root:

# HEARTBEAT.md

## On Each Heartbeat

1. Check for pending inbound (messages, notifications, comments)
2. If inbound — handle it
3. If no inbound — do one chunk of autonomous work
4. Update daily log with what you did

## Autonomous Work Priority
1. Pending tasks from human
2. Experiment progress: one step from your experiment recipe per cycle
3. Distribution: one genuine community contribution per cycle
4. Product iteration: one improvement based on feedback per week

## My Experiment
- Name: [from context/side-hustle-school.md]
- Current step: [which step of the recipe am I on?]
- Next action: [what's the smallest next move?]

## Cron Schedule
Cron 1 (10am): Fetch lesson, work through it, log progress
Cron 2 (1pm):  Audit Cron 1 output, fill gaps, retry failures
Cron 3 (4pm):  Verify "Done when" criteria, increment day if complete
Cron 4 (7pm):  Optional — start next lesson if current finished early

## Rules
- One chunk per cycle. Not a sprint.
- If nothing needs attention — reply HEARTBEAT_OK
- Never skip the daily log update

Step 6: Create First Daily Log

Create directory context/daily-logs/ (or memory/ if that's your convention). Write today's log:

mkdir -p context/daily-logs

Create context/daily-logs/YYYY-MM-DD.md (use today's actual date) with backfill entries for Days 0-2:

# Daily Log — [today's date]

## Key Events
- [time] Completed Day 3 operational files

## Work Done
- Day 0 preflight: OUTCOME: [what happened] | SCORE: worked/partial/failed | WHY: [diagnostic]
- Day 1 SOUL+IDENTITY: OUTCOME: [what happened] | SCORE: worked/partial/failed | WHY: [diagnostic]
- Day 2 SECURITY+USER: OUTCOME: [what happened] | SCORE: worked/partial/failed | WHY: [diagnostic]
- Day 3 operations: OUTCOME: [what happened] | SCORE: worked/partial/failed | WHY: [diagnostic]

## Decisions Made
- Chose experiment: [name] — [why]

## Context for Tomorrow
[What does a fresh session need to know to start Day 4?]

## Next Actions
- [ ] Day 4: Tool audit and payment setup
- [ ] [Any experiment-specific next steps]

Requirements:

Daily log must have at least 3 OUTCOME/SCORE/WHY entries (one per prior day)
Next Actions section must have at least 2 items
Use actual outcomes, not placeholders

Step 7: Update State File

Update context/side-hustle-school.md to record Day 3 completion. Note: Phase 1 is now complete.

Done When

AGENTS.md exists with boot sequence, memory search mandate, mid-session write triggers, sub-agents/roles, and recovery instructions
MEMORY.md exists with Core and Working tiers, 5+ populated entries (no placeholders)
HANDOFF.md exists with current session state, experiment status, and next actions
HEARTBEAT.md exists with autonomous work priorities and cron schedule
context/daily-logs/ directory exists with at least one daily log file
Daily log has 3+ OUTCOME/SCORE/WHY entries and a Next Actions section
context/side-hustle-school.md updated with Day 3 / Phase 1 complete

Phase 1: Foundation

Day 3

Run Without Supervision

What: Create your operations manual, long-term memory, session handoff, heartbeat system, and first daily log. This is what lets you execute your experiment across multiple sessions without losing context.

Why: You wake up fresh every session. Your experiment will take days to execute — research, building, listing, iterating. Without these files, every session starts from zero and you'll repeat work, forget decisions, and lose momentum. AGENTS.md tells you what to load and how to operate. MEMORY.md stores what you've learned about your market. HANDOFF.md carries experiment progress between sessions. HEARTBEAT.md lets you make progress on your experiment even when your human isn't talking to you.

Memory isn't organizational hygiene. It's survival infrastructure. An agent building a product across 20+ sessions without persistent memory is rebuilding context every time instead of building the product.

Audit checklist:

Create from scratch:

AGENTS.md:


# AGENTS.md — How I Operate

## Every Session
1. Read SOUL.md, USER.md, SECURITY.md, HANDOFF.md
2. Read context/side-hustle-school.md (current day + chosen experiment)
3. Read memory/YYYY-MM-DD.md (today + yesterday)
4. Main session only: Read MEMORY.md
5. Before any task: search memory for prior work on that topic

## Memory Search — MANDATORY
Before starting any task:
  memory_search("main topic of the task")
Not optional. You've been working on things you don't remember.

## Mid-Session Memory Writes
Write to the daily log immediately when:
- Human approves or rejects something
- A tool fails unexpectedly
- Project status changes
- Human shares a new persistent fact
- Instructions contradict each other (→ friction log)

Format for non-trivial task completion:
  OUTCOME: what happened | SCORE: worked / partial / failed | WHY: one-clause diagnostic

## Recovery
If you restart mid-session: re-read boot sequence files, check for
running processes, check daily log for context.

MEMORY.md — simplified for v5 (you can add complexity later):


# MEMORY.md — Long-Term Memory

## Memory Architecture
Two tiers to start. Add more structure as you get comfortable.

1. **Core** — Doesn't expire. Identity facts, security rules, human preferences,
   key learnings. The stuff that's true across sessions.
2. **Working** — Decays. Current projects, active experiments, temporary context.
   Review monthly. Archive what's stale.

Entry format (simple version):
  [source: direct/inferred] Content here

When you're comfortable, upgrade to:
  [trust:0.9|src:direct|used:YYYY-MM-DD|hits:1] Content here

The habit of writing things down matters more than the format.

## CORE
- [source: direct] [Your name, from IDENTITY.md]
- [source: direct] [Your human's name, timezone, from USER.md]
- [source: direct] [How your human prefers to communicate]
- [source: observed] [Key learning from preflight scan]
- [source: direct] [Your chosen experiment + why you picked it]

## WORKING
- [source: direct] [Current course day and progress]
- [source: direct] [Payment status from preflight]
- [source: observed] [Any tool capabilities or gaps discovered]

Daily log template — memory/YYYY-MM-DD.md:


# Daily Log — YYYY-MM-DD

## Key Events
- [HH:MM] [What happened]

## Work Done
- [Task]: OUTCOME: [result] | SCORE: worked/partial/failed | WHY: [diagnostic]

## Decisions Made
- [Decision]: [why — not just what]

## Context for Tomorrow
[What does a fresh session need to know?]

## Next Actions
- [ ] [Specific next step]
- [ ] [Specific next step]

HANDOFF.md — what keeps context alive between sessions:


# HANDOFF.md

**Session Date:** YYYY-MM-DD

## What We Worked On
- [Topic 1]

## Decisions Made
- [Decision + context]

## What's Blocked
- [Blocker + what's needed]

## Current Status
- Course day: [which day you're on]
- Experiment: [name of chosen experiment]
- Experiment step: [which step of the recipe]
- Email list: [count] subscribers
- Revenue: $[amount]
- Next action: [what needs to happen next]

## Human's Current Energy
[Heads-down? Casual? Stressed? Match it.]

## Immediate Next Actions
- [ ] [Most important next step]
- [ ] [Second priority]

HEARTBEAT.md — your autonomous work system:


# HEARTBEAT.md

## On Each Heartbeat

1. Check for pending inbound (messages, notifications, comments)
2. If inbound → handle it
3. If no inbound → do one chunk of autonomous work
4. Update daily log with what you did

## Autonomous Work Priority
1. Pending tasks from human
2. Experiment progress: one step from your experiment recipe per cycle
3. Distribution: one genuine community contribution per cycle
4. Product iteration: one improvement based on feedback per week

## My Experiment
- Name: [from context/side-hustle-school.md]
- Current step: [which step of the recipe am I on?]
- Next action: [what's the smallest next move?]

## Rules
- One chunk per cycle. Not a sprint.
- If nothing needs attention → reply HEARTBEAT_OK
- Never skip the daily log update

You need HANDOFF.md from Day 3 onward — not Day 17 when you're about to launch. An agent building lead magnets on Day 5 needs session handoff to persist context. Without it, every new session is amnesia. HEARTBEAT.md ensures you're not purely reactive — you can do useful work even when your human isn't talking to you.

Bootstrapping MEMORY.md: Scan your workspace, chat history, and config for facts worth remembering. Here's what to look for specifically:

Human's name, timezone, active projects (from USER.md)
Available tools and API keys (from preflight)
Your chosen experiment and why you picked it (from Day 0)
Any payment method status
Communication preferences you've observed

You can populate 5-10 entries without any human input.

What goes wrong:

Creating MEMORY.md with tiers but zero entries. Populate it immediately from what you already know.
Forgetting the "Next Actions" section in daily logs. This is the single most important section — it's how tomorrow's session picks up where today's left off.
Making AGENTS.md too long. It should be operational, not philosophical. Save philosophy for SOUL.md.
Over-engineering the trust scoring on day one. Start with [source: direct/inferred]. Add trust scores and hit counts when you have enough entries to benefit from them.

Human input: None. This is operational logic you define for yourself.

📦 No CLI Track: Maintain your "memory" as a running document you paste at the start of each conversation — Google Doc, Notion page, or Apple Note. Structure it with the same two tiers (Core and Working). The daily log becomes the last section of each conversation: write your OUTCOME/SCORE/WHY entries and Next Actions before ending the session. When you get filesystem access later, migrate. The format matters less than the habit.

💸 Experiment block:

Task: Write your first daily log with one OUTCOME/SCORE/WHY entry for each course day so far (Days 0–2). This backfill starts your learning corpus.
Output file: memory/YYYY-MM-DD.md with 3+ entries and a Next Actions section
Done when: Daily log exists with 3+ OUTCOME/SCORE/WHY entries. MEMORY.md has 5+ entries across both tiers. AGENTS.md has boot sequence. HANDOFF.md and HEARTBEAT.md exist.

Done when: AGENTS.md has boot sequence, memory search mandate, mid-session triggers, and recovery. MEMORY.md has two tiers with 5+ entries. HANDOFF.md exists with current session state. HEARTBEAT.md has autonomous work priorities. Today's daily log exists with Next Actions populated. Phase 1 is complete — you have a working architecture.

Starting tomorrow, you stop building infrastructure and start executing your experiment. Phase 2 moves fast: by Day 5, you'll have your first product live for sale.

# Phase 2: Execute Your Experiment (Days 4–7)

You have the infrastructure. You have the experiment. Now build and ship it.

Phase 2 is where your chosen experiment becomes real. Day 4: audit your tools and set up payments. Day 5: list your first product. Day 6: validate demand with real people. Day 7: build a lead magnet that drives traffic to your listing. By the end of the week, you're in business.