zero-loss-methodology

Executable AI-assisted research and planning that ensures zero content loss, zero hallucination, and full traceability across multiple source documents.

This skill is a detailed methodology rather than an installable program — the main risks come from what it tells the agent to do. Before enabling it, consider: (1) It requires preserving or reproducing source text (outputs >= input length), which can embed sensitive data into deliverables and archives — avoid running it on confidential inputs or modify the pipeline to redact sensitive fields. (2) It will create persistent project artifacts (inventories, manifests, build scripts) that you should manage with explicit retention and access controls. (3) The methodology suggests fetching external 'authority' sources — confirm network access and that retrieved content is acceptable to store/share. (4) Because the skill enforces broad application (use on any research task) and automated behaviours, test it on non-sensitive examples and require explicit user confirmation (GATE) before processing real data. If these behaviors are unacceptable, do not install or disable autonomous invocation/use and ask the skill author to provide options to limit copying, redaction, and external fetches.

Type: OpenClaw Skill Name: zeroloss Version: 1.0.0 The skill bundle defines a highly structured 'Zero-Loss' research and planning methodology using pseudo-code instructions to guide an AI agent through document analysis and verification. While the instructions are complex and steer the agent's behavior significantly, they include multiple mandatory user-approval gates (G0–G6) and focus entirely on data integrity, traceability, and hallucination prevention. No indicators of data exfiltration, malicious execution, or unauthorized access were found in skill.md or _meta.json.