The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill recommends the installation and execution of code from unverified third-party sources.
In references/imessage-video-ads.md, it suggests using npx gooseworks install --all to install an external pipeline. This repository is noted as public but lacks an open-source license.
In references/generative-tools.md, it instructs the user or agent to clone and setup an external project: git clone https://github.com/jamiepine/voicebox.git followed by make setup and make dev.
[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves processing untrusted external data, which could contain malicious instructions.
Ingestion points: Reads customer reviews and campaign comments from the inputs/reviews/ and inputs/comments/ directories (SKILL.md).
Boundary markers: The instructions lack explicit delimiters or warnings to the agent to treat this data as untrusted content when interpolating it into prompts.
Capability inventory: The skill uses subprocess execution to interact with google-ads, meta-ads, linkedin-ads, and tiktok-ads CLI tools, and performs file writing in dated output folders.
Sanitization: There is no mention of sanitizing or validating the ingested reviews or comments; instead, the skill emphasizes using the customer language "verbatim" to ensure grounded generation.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands to pull performance data and manage campaigns.
Evidence includes example commands like node tools/clis/google-ads.js reports get and references to using various platform CLIs (Google Ads, Meta, LinkedIn, TikTok) in SKILL.md and references/creative-roadmap.md.
[DYNAMIC_EXECUTION]: The skill describes a complex pipeline for generating video ads that involves rendering HTML/CSS/JS in a headless browser (Playwright) driven by JSON data scripts (references/imessage-video-ads.md). While this is a common approach for the described task, execution of runtime-generated content on a local system carries inherent risks.

ad-creative