Agent OS for Solo Developers: Skills, Memory, Jobs, Handoff

You do not need a 40-person platform team to run an agent OS.

You need a small operating system for your work: skills, memory, jobs, and handoff - with human gates on anything irreversible.

This is the solo-developer version of the patterns behind the Agent OS checklist and the Agent OS consulting package.

Kill the default approach

Default: more agents, more cron, no handoff, no green command.

What breaks first: 3am half-jobs, Monday amnesia, a hobby only you understand.

By the end you will have

1. The four solo pieces (skills, memory, jobs, handoff)
2. The Weekend Agent OS Layout (magnet)
3. A one-page scorecard
4. Solo anti-patterns
5. Path options (checklist / OSS / package)

The problem solo builders hit

Chat agents are great until:

• Monday you forget Friday's decisions
• A cron half-works at 3am and you have no board
• You cannot hand a repo to a collaborator without a call
• The agent "helps" by rewriting the wrong module again

That is not a model gap. That is a missing OS.

The four solo pieces

1. Skills (behavior contracts)

Markdown contracts the agent loads for repeated work. See the skills beginner guide.

Solo rule: max 10 skills until each has been used twice.

2. Memory (journals, not dumps)

Write short receipts:

• Decision
• Evidence
• Next action

Tools: project docs, Obsidian, or a simple journal/ folder. Avoid pasting entire chats into the next session.

3. Jobs (scheduled work with gates)

Cron is useful when:

• The task is boring and frequent
• Output is reviewable
• Failure is visible

Cron is dangerous when it publishes, deploys, or spends without you.

Solo pattern: cron drafts, human approves.

4. Handoff (so future-you is a teammate)

A handoff page answers:

• How to run the project
• How to run the agent
• What not to touch
• Where logs and journals live

If only you understand it, you do not have an OS. You have a hobby.

Magnet: Weekend Agent OS Layout

Name: Weekend Agent OS Layout

repo/
  AGENTS.md          # how agents should behave here
  skills/            # 3-10 contracts
  journal/           # short dated notes
  scripts/check.sh   # one command that means "green"

Acceptance:

1. ./scripts/check.sh is the only "are we good?" entry
2. An agent can run it
3. A stranger can read AGENTS.md and not destroy prod

You should see: green command runs from the agent, and a handoff page a collaborator can follow without a call.

Where open source helps

• Mission Control - fleet / job visibility
• awesome-hermes-agent - Hermes map
• LACP - local control-plane patterns

Install what you will operate. Ignore the rest.

Failure modes

One-page scorecard

Score 0-2 each: skills used · journals last 7 days · gated job · handoff doc · green command.

0-4: chatting. 5-7: AI-native solo. 8-10: ready for a second human or agent.

When not to build an agent OS yet

• You have never finished a PR with one agent
• You have no green command
• You want a toy demo, not a compounding loop

Path options

Bottom line

Build the OS small. Then let it compound.

Your next action: scaffold Weekend Agent OS Layout in one repo and make scripts/check.sh green once.