The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and analyze untrusted performance data from external sources.
Ingestion points: Performance data provided via CSV, API output, or text paste as described in Mode 2 of SKILL.md and evaluation ID 3 in evals/evals.json.
Boundary markers: None; there are no delimiters or specific instructions to the agent to ignore potentially malicious commands embedded within the performance data.
Capability inventory: The skill generates new ad variations and strategy recommendations based on the input; it also suggests executing external CLI tools to fetch this data.
Sanitization: None; the skill does not provide instructions for escaping or validating the content of the ingested metrics.
[COMMAND_EXECUTION]: The skill provides instructions and examples for executing local commands and scripts.
SKILL.md contains a workflow example that executes a local Node.js script (node tools/clis/google-ads.js) to retrieve ad performance reports.
references/generative-tools.md provides examples for using ffmpeg to merge audio and video files.
[EXTERNAL_DOWNLOADS]: The skill references the download and installation of external software and repositories.
Includes instructions for cloning a repository from GitHub (github.com/jamiepine/voicebox).
References the use of npx to initialize Remotion video projects (npx create-video@latest).
These references target well-known developers and established technology platforms.

ad-creative