agentafk NEW

agentafk is an open-source unattended runtime for coding agents that enforces permission boundaries, explicit terminal states, and local trace history.It orchestrates subagents and runs parallel hypotheses in isolated worktrees, recording every tool call and session trace to ~/.

afk/ for reproducibility and audit.Verification workflows include adversarial re-derivation (/shadow-verify) and automated checks to confirm agent outputs before irreversible actions.Reversibility rules and gated operations protect larger-impact actions (commits, pushes, deletes) while allowing reversible edits and scratch files.

Composable skills and commands (/mint, /forge, /diagnose, /parallelize, /review) enable end-to-end pipelines from spec to ship with built-in verification steps.Remote supervision via Telegram, interrupt/resume controls, and on-disk session state support monitoring, recovery, and rollback of long-running runs.

Designed for developer teams using Claude, Cursor, or other coding agents that require unattended execution, verifiable results, audit trails, and safe rollbacks.

Key Features

• Unattended runtime enforcing permission boundaries, explicit terminal states, and local trace history
• Orchestrates subagents and runs parallel hypotheses in isolated worktrees, recording every tool call and session trace to ~/.afk/ for reproducibility and audit
• Verification workflows including adversarial re-derivation (/shadow-verify) and automated checks to confirm agent outputs before irreversible actions
• Reversibility rules and gated operations protecting high-impact actions (commits, pushes, deletes) while allowing reversible edits and scratch files
• Composable skills and commands (/mint, /forge, /diagnose, /parallelize, /review) plus remote supervision via Telegram, interrupt/resume controls, and on-disk session state for monitoring, recovery, and rollback

Use Cases

• Automate large-scale codebase refactors and dependency updates using agent afk's unattended runtime with isolated worktrees and permissioned execution, while session trace auditing and gated rollback ensure changes are reproducible, verifiable, and safely reversible without manual oversight
• Run autonomous CI/CD maintenance tasks and security patching across multiple repositories using agent afk's composable agent workflows and parallel hypotheses, enforcing adversarial output verification and permission boundaries to prevent unsafe deployments and record full session/tool-call traces for auditability
• Orchestrate complex developer operations like database schema migrations, feature-branch testing, and multi-step releases with agent afk's gated reversibility and isolated worktree orchestration, enabling safe experimentation, automatic verification, and traceable rollbacks for incident response and compliance

Who is it for?

• Coding developers
• Autonomous software engineers
• Open-source maintainers
• Engineering managers
• Mlops engineers