AI Agent Financial Observability — monitor, budget, and analyze spending across any AI agent. Track costs, set budgets, detect anomalies, and export metrics...

This skill appears to be what it claims (an on-agent financial observability SDK), but review the following before installing or enabling it: - Network exposure: the Dashboard (and Prometheus exporter) start HTTP servers and the Dashboard example binds to 0.0.0.0 by default. If you run this on a machine with network access, lock it down (bind to localhost, use a firewall, or put behind authenticated proxy). - Data exfiltration: WebhookExporter will POST every transaction to the configured URL. Only configure trusted endpoints and avoid embedding sensitive API keys in code or in public repos. Prefer the Console or local Jsonl exporters for testing. - Local persistence: transactions are written to JSONL on disk by default. Ensure the persist path is acceptable, has correct permissions, and does not leak to backups or shared volumes. - Dependency/runtime: Prometheus and httpx are optional and imported at runtime; installing optional extras will add those dependencies. Review the package on PyPI or its source repository before pip installing, especially since the skill owner/source are not provided in the metadata. - Least privilege and isolation: run the tool in an isolated environment (container, VM, or sandbox) until you confirm configuration and network exposure are safe. If you want a tighter posture, ask the maintainer (or inspect the SpendTracker/tracker implementation in full) for the defaults used for persistence paths, dashboard host binding, and any automatic exporter defaults; change them to local-only, authenticated, or disabled as appropriate.

Type: OpenClaw Skill Name: agentfinobs Version: 1.0.0 The skill is classified as suspicious due to multiple high-risk capabilities and vulnerabilities, primarily related to arbitrary file I/O and data exfiltration capabilities. The `agentfinobs/__main__.py` CLI command `status` allows reading from a user-specified JSONL file, which could be exploited for arbitrary file disclosure. The `agentfinobs/tracker.py` and `agentfinobs/exporters.py` (JsonlExporter) allow writing transaction data to user-specified file paths, posing an arbitrary file write vulnerability. Furthermore, the `agentfinobs/exporters.py` (WebhookExporter) provides a direct capability to exfiltrate transaction data (including potentially sensitive descriptions and tags) to any configurable external HTTP endpoint. While these features are intended for legitimate observability and integration, they present significant attack surfaces for prompt injection or misuse, allowing an attacker to read/write arbitrary files or exfiltrate data to malicious destinations if the agent's input can be controlled.