Parallax

Views every decision from five simultaneous stakeholder perspectives — the user who touches it, the developer who maintains it, the operator who deploys it,...

This skill appears coherent and low-risk because it is instruction-only and declares no credentials or installs. Before installing, open and read the full SKILL.md (the provided excerpt was truncated) to confirm it does not: (1) instruct the agent to read system files or shell history, (2) request or reference environment variables or secret tokens, or (3) direct data to external endpoints. If you plan to let agents call skills autonomously, restrict which tools the agent can use (file access, shell, network) or run it in a sandboxed environment so the skill cannot access secrets or external systems without your explicit approval.

Type: OpenClaw Skill Name: parallax Version: 1.0.0 The OpenClaw skill bundle 'parallax' is benign. It provides a conceptual framework for multi-perspective decision-making, instructing the AI agent on how to analyze proposed changes from five different viewpoints (User, Maintainer, Operator, Adversary, Sponsor). The `SKILL.md` explicitly states 'Zero external dependencies. Zero API calls.' and `requires_api: false` in its metadata, confirming it does not perform any network requests, file system operations beyond its own context, or execute external commands. All instructions are for the AI's cognitive process and output formatting, with no evidence of prompt injection for malicious purposes, data exfiltration, or unauthorized actions.