longrunning-agent_skill

Overview

This skill enables AI agents to run long-running projects across multiple sessions with reliable continuity. It provides structured task tracking, priority and dependency management, and a simple session-based progress log to drive incremental, testable work.

How this skill works

The agent inspects a project directory for a workflow set: a human-readable project guide, a task list (JSON) with priorities and dependencies, and a progress log. Each session the agent reads progress, picks the next unmet task whose dependencies are satisfied, runs any initialization, makes a small atomic change, tests, documents the result, and marks the task complete. Commits are kept per-task to preserve atomic history.

When to use it

• When projects span hours, days, or weeks and need persistent state between sessions.
• For complex features that must be decomposed into dependent, testable steps.
• When multiple agents or humans may resume work and need clear handoff state.
• To enforce single-task-per-session discipline for reliable progress tracking.
• When reproducible, atomic commits are required for auditability.

Best practices

• Work on exactly one task per session to keep changes atomic and reviewable.
• Maintain concise progress entries with timestamps and outcomes.
• Define explicit dependencies in the task JSON to enforce order.
• Run linting, builds, and tests before marking a task as passed.
• Commit after each completed task with a clear message referencing the task id.

Example use cases

• Incrementally implementing a new product feature split into dependent subtasks.
• Refactoring a large codebase where each session applies and tests a single refactor step.
• Documenting and delivering research or writing projects across multiple sittings.
• Coordinating multi-agent workflows where agents pick next available tasks automatically.
• Maintaining a reproducible backup or archive workflow with logged progress and commits.