Xiaguang Harness [DEPRECATED → use trinity-harness]

Production-grade Agent Harness combining execution discipline (Superpower), knowledge compounding (CE), and product thinking (Gstack) into a single adaptive...

This skill is an instruction-only engineering harness and appears to be what it claims. Before installing or enabling it for an agent, confirm whether you are comfortable granting the agent: (1) filesystem write/read access (it will create checkpoints and overwrite files), (2) the ability to run shell/test/CI commands (ls, grep, wc, test runners, git), and (3) the ability to interact with or restart services. The metadata lacks declared required binaries or owner/homepage and the skill is marked deprecated in its name — prefer the recommended replacement (trinity-harness) if available. If you will run this on production systems, restrict the agent's permissions (run in an isolated repo/container, limit paths, require human confirmations for deploy/restart steps) and audit any commits or service restarts the agent proposes.

Type: OpenClaw Skill Name: xiaguang-harness Version: 1.0.1 The skill bundle defines a structured engineering workflow (Agent Harness) designed to improve AI agent reliability through task grading, verification protocols, and checkpointing. The instructions in SKILL.md and the reference files (checkpoint-patterns.md, mast-failure-taxonomy.md) focus entirely on software development best practices and contain no evidence of malicious intent, data exfiltration, or unauthorized execution.