Security audit
Security checks for vulnerabilities and agentic risk
This Meta Ads monitoring skill is coherent, but it needs review because its example API calls place the access token in the URL where it may leak through logs or diagnostics.
Review before installing. Use only a least-privilege, read-only Meta token, avoid shell tracing or verbose logs, and prefer changing the examples to send the token with an Authorization header or a credential-aware SDK instead of a URL query parameter.
SKILL.md:60OAuth Access Token Exposed in URL Query Parameters
The description says to use the skill 'every morning or before any daily review meeting,' which describes broad situations rather than a narrowly defined activation trigger. In a markdown skill file, this lack of specific trigger scope can cause unintended invocation in many ordinary review or meeting-prep contexts.
The section instructs to use the skill 'every morning or before any daily review meeting' and frames it around a broad question, but it does not clearly distinguish this skill from other common monitoring or reporting requests. Without explicit exclusions or trigger constraints, activation criteria remain vague for a markdown skill description.
The skill explicitly instructs the agent to transmit advertising performance data and a bearer access token to an external service via curl. Although the destination is the legitimate Meta Graph API and the business purpose is expected, this is still an external data transmission path involving sensitive credentials and account analytics, so misuse, over-collection, logging leakage, or invocation in the wrong context could expose confidential data.
Use exec + curl to call the Meta Graph API:
Account-level (yesterday):
The documentation explicitly states this is a detection-only tool and says not to output recommendations, routing action plans to another skill instead. However, the formatting and rules sections require 'Immediate action' and 'One immediate action per issue', which directly contradicts the stated boundary.