Self Discipline

Guarantee instruction compliance with root cause analysis, flow verification, and automated validators that make future failures impossible.

This skill is internally consistent with its stated purpose, but it will create files and executable validators in your home directory and may propose edits to AGENTS.md or heartbeat files so rules are actually loaded. Before installing or granting permission: 1) Confirm you want ~/self-discipline/ created and review the exact files the skill will write. 2) Inspect any validator scripts (pre-send, pre-commit, pre-action) before making them executable — they run locally and can block actions; ensure their behavior is acceptable. 3) If the skill requests edits to AGENTS.md/HEARTBEAT.md, ask to see the exact diffs and make backups. 4) Keep validators under version control and test them manually first. 5) If you don’t want the skill to influence other agent sessions, decline edits to AGENTS.md/heartbeat or keep activation_mode set to manual. If you want additional assurance, paste the actual validator scripts here and ask for a line-by-line review.

Type: OpenClaw Skill Name: self-discipline Version: 1.0.1 The skill is designed to enforce agent discipline by generating and executing shell scripts (validators) and modifying system-level configuration files (e.g., Git hooks, HEARTBEAT.md). While the skill's documentation (SKILL.md, setup.md, validators.md) explicitly emphasizes user consent, transparency, and security best practices (e.g., validators never modify data, no external network requests, no credential access), the inherent capability to generate and execute arbitrary shell code based on agent instructions, and to modify critical system integration points, introduces a significant risk. This powerful capability, even with stated good intent, makes the skill 'suspicious' due to the potential for prompt injection or agent misinterpretation leading to vulnerabilities, despite the skill's internal safeguards.