Away Mode

Development stalls when you are away from the keyboard. This skill allows your agent to work through task lists autonomously while you are asleep or in meetings, making decisions on its own and providing verifiable proof for every completed job.

What it does

• Autonomous execution completes a specific list of tasks from your todo.md or backlog without asking for clarification.
• State persistence maintains a progress.json file to ensure the agent can resume exactly where it left off if the session drops.
• Verification loops runs actual tests, database queries or API calls to prove changes are live rather than just claiming they are done.
• Automated decisioning resolves ambiguities by following existing project patterns and logging assumptions instead of pausing for user input.
• Detailed handoff generates a HANDOFF.md report with status tables, command output evidence and a list of pending approvals for irreversible actions.

How it works

1. The user provides a task file or a list of jobs and signals they are going "away" or "offline."
2. The agent initializes a tracking folder, reads project conventions (README, CLAUDE.md), and maps the verification targets.
3. The agent enters a loop: planning, applying file changes, verifying on the server/database, and recording evidence.
4. If a session interruption is likely, the agent schedules resume triggers to ensure the entire backlog is finished.

Frameworks & tools

Works with any stack that includes a task file (.md, Notion, Drive) and a command-line interface. It specifically integrates with docker-compose, prisma, alembic, and standard Unix tools for file manipulation and verification.

Why this beats prompting it yourself

Standard prompts often stall when they hit an error or an ambiguity, waiting for your reply. This skill enforces a "decide and record" policy and uses persistent state tracking to survive session timeouts, turning your agent into a true overnight worker.

Use cases

• Overnight refactoring to handle repetitive migrations or style updates across a large codebase.
• Unattended deployments where the agent must verify container status and API health after a series of updates.
• Syncing backlogs by working through a sprint.md file while the lead developer is in meetings.

Known limitations

Does not perform irreversible actions like dropping production tables or sending external emails without explicit prior permission. Requires access to a local filesystem or device bridge to maintain the state file.