long-horizon-workflows

What problem does it solve?

Long-horizon agent work often fails mid-execution without recovery, making multi-hour plans unreliable and difficult to audit or resume. This Skill provides a practical blueprint for breaking objectives into staged phases with checkpointing, progress visibility, human approval gates, and error recovery so work can continue after interruptions.

Core Features & Use Cases

• Multi-phase pipeline orchestration: Decomposes an objective into ordered phases (data collection, analysis, recommendations, reporting) to match real production workloads.
• Checkpointing for resumability: Persists intermediate state between phases so execution can resume safely instead of restarting from scratch.
• Human-in-the-loop gates for safety: Pauses before high-stakes decisions (e.g., budget or bid changes) to prevent catastrophic unattended actions.
• Progress tracking and auditability: Tracks phase status, estimates completion progress, and produces an execution trail for debugging and compliance.
• Error handling with recovery policy: Applies per-phase and per-task behaviors (retry/skip/escalate/abort) to survive partial failures.

Quick Start

Ask an AI to design a long-horizon workflow for your next multi-stage Google Ads account audit with checkpoints after each phase and a human approval gate before any irreversible optimization changes.