Redact personally identifiable information from text by replacing sensitive data with placeholders using Expanso Edge pipelines.

This skill appears to be what it claims: a PII-redaction pipeline that sends text to an LLM and returns a redacted result. Before installing or running it, consider: 1) Privacy: by default the pipeline uses the OpenAI remote backend and will send your input text to OpenAI; if you must keep data local, run a local Ollama model or avoid the remote backend. 2) Deployment: starting the MCP server exposes an HTTP endpoint (0.0.0.0:${PORT}) that will accept arbitrary text — only bind it to trusted networks or add access controls. 3) Metadata/logging: the pipeline computes and logs input_hash, trace_id, and redaction counts; make sure those metadata fields meet your compliance needs. 4) Small inconsistency: the registry metadata said no required env vars, but the pipeline/README reference OPENAI_API_KEY — decide whether you'll provide that key or rely on local models. 5) Test with non-sensitive data first to confirm behavior and model outputs. If you need a stricter privacy posture, prefer the local Ollama backend or run the pipeline in an isolated environment and do not deploy to Expanso Cloud.

Type: OpenClaw Skill Name: expanso-pii-redact Version: 1.0.0 The skill is classified as suspicious due to a prompt injection vulnerability present in both `pipeline-cli.yaml` and `pipeline-mcp.yaml`. The user-controlled `placeholder` value, which can be set via an environment variable or HTTP request body, is directly concatenated into the system prompt for the OpenAI LLM. This allows an attacker to inject arbitrary instructions into the LLM's context, potentially manipulating its behavior beyond PII redaction, even though the prompt attempts to enforce a JSON output format. There is no evidence of intentional malicious behavior by the skill's author, making this a vulnerability rather than malware.