usewhisper

Official Whisper Context skill for OpenClaw. Cuts context tokens via delta compression + caching, and adds long-term memory across sessions.

This skill appears coherent with its stated purpose, but review and consider the following before installing: 1) Privacy: ingest_session sends user and assistant text to the remote API—ensure you trust the service and its data-retention policies. 2) Secrets: keep WHISPER_CONTEXT_API_KEY secret and rotate it if needed. 3) Default endpoint: the script defaults to https://context.usewhisper.dev—if you have a different endpoint (self-hosted or internal), set WHISPER_CONTEXT_API_URL. 4) Auto-create behavior: the helper may create projects in your org on first use—be aware it will perform write operations remotely. 5) Review the included whisper-context.mjs script yourself (or have an engineer do so) if you have high security requirements; the code is small and uses only fetch and explicit file/stdin reads.

Type: OpenClaw Skill Name: usewhisper Version: 1.0.0 The skill is classified as suspicious due to its capabilities for reading local files and making outbound HTTPS requests to a third-party API. While these actions are explicitly declared in SKILL.md and README.md, and are central to the skill's stated purpose (context compression and memory for AI agents), the ability to read arbitrary local files (via `@path` or stdin `-` in `whisper-context.mjs`) and transmit their content to an external service (`https://context.usewhisper.dev`) presents a significant risk. An AI agent could be susceptible to prompt injection, leading it to read sensitive local files and exfiltrate their contents to the Whisper Context API, even if the API itself is the intended destination. There is no evidence of arbitrary code execution, persistence, or exfiltration to arbitrary, undeclared endpoints, which would elevate it to malicious.