Agent Harness — Multi-session Project Framework

What it does

Agent Harness organizes multi-session development workflows: sprint planning, single-feature coding sessions, and parallel subagent orchestration. The skill documents slash-command routing and scripts to initialize projects, track features, and archive sprints so the agent can manage project state across sessions and automate routine project-maintenance tasks.

When to use it

Use when a user requests help breaking requirements into features, running sprint planning, starting a coding session that follows a one-feature-per-session protocol, or executing parallel work with subagents (the 996 pattern). Trigger phrases include: "sprint planning", "init project", "start coding session", "parallel subagents".

What's included

• Scripts: The repository includes scripts for init_project, status checks, resolve_project_dir, validate_structure, and archive utilities (all callable via python3).
• References: The SKILL.md points to detailed reference docs for sprint-agent, coding-agent, and 996-agent workflows.
• Instructions: The skill provides explicit command syntax, project file rules (features.json, progress.md), session protocols (one feature per session), and critical rules for commits and progress logging.

Compatible agents

Best for agents that orchestrate multi-step workflows and interact with repositories and shells (Claude Code and other coding-focused agents). It is a governance/coordination skill rather than a UI component.