Skip to content

Solo Harness: Give Your Agent a Task, Go To Sleep, Check Results In The Morning

Coding agents are great at short tasks. "Fix this bug", "write this function", "refactor this file" — done in minutes.

But what about bigger objectives? "Refactor the entire auth module and write tests." That's hours of work. You can't sit there and watch.

Solo Harness lets you hand off a big task and come back when it's done.

The Old Way vs. The New Way

Before: You run a long task, go to bed, wake up to... something. Maybe it finished. Maybe it got stuck. Maybe it made a mess of your git history. You have no idea what happened.

After: You start a harness run with --worktree, go to bed, and in the morning:

  • Check the run journal to see exactly what happened
  • Review the structured status to see if it completed
  • If it got stuck, resume from where it stopped
  • If it went off track, delete the worktree — your main branch is untouched

How It Works

# Evening: start the run
aios harness run \
  --objective "Refactor the auth module and write integration tests" \
  --session nightly-auth \
  --worktree

# Morning: check what happened
aios harness status --session nightly-auth --json

# If it finished: review the changes
aios hud --session nightly-auth

# If it got stuck: fix the issue and resume
aios harness resume --session nightly-auth --max-iterations 10

Why --worktree Matters

The --worktree flag is important. It creates a separate copy of your repo where the agent can make changes freely.

  • Good results? Merge the worktree into your main branch
  • Bad results? Just delete the worktree — zero impact on your code

No git reset --hard. No risky cleanup. Just safe isolation.

What Gets Recorded

Every harness run writes a journal:

memory/context-db/sessions/<session-id>/artifacts/solo-harness/
  ├── objective.md           # What you asked it to do
  ├── run-summary.json       # Current state and progress
  ├── control.json           # Stop requests and notes
  ├── iteration-0001.json    # What happened in each iteration
  └── iteration-0001.log     # Detailed logs

This gives you a readable handoff trail — not just "the agent ran for a while", but exactly what it did, what worked, and what didn't.

When To Use Solo Harness

Use it when: - You have one clear objective that will take a long time - The task doesn't need to be split across agents - You want to wake up to results instead of babysitting

Don't use it when: - The task can be split into independent parts (use Agent Team instead) - You're still figuring out the requirements (start with a normal session) - You need staged execution with quality gates (use orchestrate)

Try It

# Start with a dry run to verify everything is set up
aios harness run \
  --objective "Write integration tests for the payment module" \
  --session test-dry \
  --worktree \
  --dry-run --json

# When you're ready, go live
aios harness run \
  --objective "Write integration tests for the payment module" \
  --session payment-tests \
  --worktree \
  --max-iterations 20

Solo Harness shipped in AIOS 1.7. Read the full docs or try it tonight.