Detect personally identifiable information (PII) in text using Expanso Edge pipelines for CLI, MCP server, or cloud deployment.

This skill runs Expanso pipelines that put your input text into an OpenAI chat completion call. If you provide an OPENAI_API_KEY, the full text will be sent to OpenAI. The registry metadata failing to declare the OpenAI credential and the README's claim that a local 'regex' mode can be used are inconsistent with the included pipeline files (which always call openai_chat_completion). Before installing or running on sensitive data: (1) Decide whether you are willing to send inputs to OpenAI; if not, avoid supplying an API key and verify the pipelines actually perform local regex detection (they currently do not). (2) Run the skill in a safe/test environment with non-sensitive inputs to observe behavior. (3) Inspect or modify the pipeline YAML to add a true local fallback (regex processors) if you need on-device-only detection. (4) Confirm you trust the expanso-edge binary and the environment where it will run. If you need clarification, ask the author to correct the skill metadata and to provide a documented regex-only pipeline.

Type: OpenClaw Skill Name: expanso-pii-detect Version: 1.0.0 The skill is designed to detect PII using an OpenAI LLM. Both `pipeline-cli.yaml` and `pipeline-mcp.yaml` construct the LLM prompt by directly concatenating user-provided input (`content()` or `this.text`) without apparent sanitization or robust prompt engineering to prevent manipulation. This makes the LLM vulnerable to prompt injection, where a malicious user could potentially bypass the LLM's instructions, extract information, or cause it to generate unintended output. While the skill uses the `OPENAI_API_KEY` environment variable, this is for its stated purpose and not for unauthorized exfiltration.