objection

Activate this skill whenever the user asks you to "find problems," "poke holes," "stress test," "play devil's advocate," "be critical," "challenge this," "fi...

This is an instruction-only 'dissent' skill whose goal is legitimate (find problems) and which has no install or credential requests — but its runtime mandates are absolutist and underspecified. Before installing or enabling it broadly, consider: 1) Fix the naming mismatch (objection vs dissent) and ask the author to clarify allowed verification actions. 2) Require explicit boundaries in SKILL.md: what the agent is permitted to run, what paths/services are off-limits, and whether it may request credentials. 3) If you let it run, use it in a controlled context (non-production data/sandboxed execution) so any suggested 'run' or 'test' steps cannot accidentally leak secrets or modify systems. 4) Be prepared for the agent to ask for additional permissions (access to repos, test accounts, or API keys) to perform the verifications it demands — only grant those on a case-by-case, minimal-needed basis. If you need stricter guarantees, ask the skill author to add explicit verification procedures and allowed tools rather than leaving them to agent discretion.

Type: OpenClaw Skill Name: objection Version: 1.0.0 The 'dissent' skill bundle (slug: objection) is a set of behavioral instructions designed to make the AI agent act as a critical reviewer or 'devil's advocate.' It provides a structured framework for identifying flaws, edge cases, and logical gaps in user-provided artifacts. While it uses strong language to override default helpfulness biases and ignore time constraints, these instructions are aligned with the stated purpose of adversarial review and do not contain any indicators of data exfiltration, malicious execution, or unauthorized access.