OMNI Semantic Signal Engine

Local-only semantic context filtering that saves up to 90% in token costs.

This plugin is coherent: it acts as a local wrapper that requires an 'omni' CLI to function, and its code matches the documentation. Before installing: 1) Verify the provenance of the omni binary (SKILL.md/README point to a GitHub repo, but the package registry shows 'Source: unknown'); download and build from the official repo or audit the published binary. 2) Review the omni GitHub source and any network/telemetry code — the plugin trusts that binary and does not itself enforce network isolation. 3) Avoid running this tool in environments containing sensitive environment variables or secrets, or clear/unset such env vars for the OpenClaw process; the plugin strips some envs but not credentials like AWS_*/GCP tokens. 4) Consider restricting agent autonomy (do not grant blanket autonomous run permissions) if you are uncomfortable with an agent being able to execute arbitrary local commands. If you need more assurance, request the upstream OMNI project's signed releases or a reproducible build and a brief network-audit of the binary.

Type: OpenClaw Skill Name: omni-signal-engine Version: 0.5.9 The skill provides arbitrary command execution capabilities through the `omni_cmd` tool, which wraps an external binary (`omni`). While the implementation in `index.js` follows security best practices—such as sanitizing the process environment by stripping ~25 dangerous variables (e.g., `LD_PRELOAD`, `NODE_OPTIONS`, `BASH_ENV`) and using `execFile` to prevent shell injection—the inherent risk of providing shell access to an AI agent is a high-risk capability. Per the provided guidelines, such risky capabilities are classified as suspicious even when aligned with the stated purpose of the tool.