Security audit
Security checks for vulnerabilities and agentic risk
This presentation-generation skill is not malicious, but it asks for broad read, write, URL, API, and command-execution capability without enough concrete boundaries.
Review before installing. Use it only on non-sensitive inputs, approve URL fetching intentionally, and do not allow arbitrary shell commands unless the agent shows the exact command and it is clearly needed to generate the presentation or PDF.
The activation guidance is broad enough that an agent may invoke the skill in loosely related file-processing or content-conversion scenarios without clear boundaries. In a skill with read, write, and especially exec capabilities, ambiguous invocation criteria increase the chance of unsafe use, over-privileged execution, or accidental processing of sensitive inputs.
The skill supports URL input but does not clearly warn that fetching or processing a user-supplied URL may transmit data to external services or contact attacker-controlled hosts. This can expose internal network metadata, trigger SSRF-like behavior in the execution environment, or surprise users who expect local-only processing.
The skill advertises that command execution is limited to a whitelist, yet only generic exec capability is declared and no enforcement mechanism is shown. This mismatch can cause downstream agents or users to trust unsafe command execution paths, enabling arbitrary shell commands if user-controlled input is ever passed into execution.
The skill claims outputs exclude sensitive data, but it also describes execution logs and API-key-based operation without showing any redaction, masking, or log filtering controls. In practice, command output, errors, environment dumps, or request traces may leak secrets into logs or generated artifacts.
No suspicious patterns detected.
