Nm Attune War Room

Convene a multi-LLM expert panel to pressure-test high-stakes decisions and build consensus through structured deliberation

This skill appears to implement a complex, legitimate multi-LLM deliberation process, but it has several surprising behaviors you should review before installing: - Credentials & CLIs: The SKILL.md assumes local CLIs (gh, gemini, qwen, claude, claude-glm/ccgd) and GitHub authentication are present. Confirm where those credentials live and that you want the skill to use them; the skill doesn't declare required env vars explicitly. - Automatic capture/publishing: By default it will (a) capture rejected COAs automatically into your local strategeion and (b) offer publishing to GitHub Discussions with publishing defaulting to "Publish" (opt-out). If you handle sensitive topics, change defaults to manual publish and require confirmation before any capture/publish. - Dangerous flags/aliases: The documentation recommends use of a '--dangerously-skip-permissions' flag and creating an alias that embeds it. Do NOT add such aliases or enable flags without understanding their effect — they bypass permission checks and are high-risk. - Inspect local scripts: Look for scripts referenced (e.g., scripts/deferred_capture.py) in the repository to verify what they do before running them. If they will be executed automatically, ensure they don't exfiltrate or modify unrelated data. - Least privilege & testing: Run the skill in a sandbox or with dummy repositories/accounts first. Verify what files it writes (the strategeion path and .attune session file) and whether the gh commands use the intended GitHub account/repo. Consider forcing explicit confirmation for any network publish or irreversible actions. If you want, provide the repository path or the contents of any referenced scripts (scripts/deferred_capture.py, any cli wrappers), and I can review those for specific risks and recommend configuration changes to make the skill safer.

Type: OpenClaw Skill Name: nm-attune-war-room Version: 1.0.0 The bundle orchestrates a multi-LLM 'War Room' using high-privilege operations, including automated GitHub Discussion publishing via 'gh api graphql' and the use of 'tmux' for agent teams. Key indicators include the explicit use of the '--dangerously-skip-permissions' flag for GLM-4.7 model integration (expert-roles.md) and the execution of an external script ('scripts/deferred_capture.py') not included in the bundle. While these capabilities support the stated purpose of strategic deliberation, the automated publishing of session context to GitHub and the bypass of security prompts represent a significant attack surface.