The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes instructions to download and build software from a personal GitHub repository, which involves executing arbitrary build commands.
Evidence: In references/generative-tools.md, the skill instructs the user to git clone https://github.com/jamiepine/voicebox.git and then run make setup && make dev. This pattern encourages the execution of unverified scripts from a third-party source.
[COMMAND_EXECUTION]: The skill makes use of local CLI tools to manage and fetch data from advertising platforms.
Evidence: SKILL.md contains examples of executing node scripts such as node tools/clis/google-ads.js reports get --type ad_performance. While this is functional, it demonstrates a dependency on executable local scripts that the agent can invoke.
[EXTERNAL_DOWNLOADS]: The skill directs users to various external websites and tools that are not part of the established trusted vendor list.
Evidence: References to voicebox.sh, klingai.com, higgsfield.ai, and seed.bytedance.com for tool acquisition and documentation.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting performance data from potentially untrusted sources to influence its output.
Ingestion points: External performance data provided via CSV files, clipboard pastes, or API outputs as described in SKILL.md under 'Mode 2: Iterate from Performance Data'.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to distinguish between data and instructions within these inputs.
Capability inventory: The skill has capabilities to read local context files (product-marketing-context.md), execute CLI tools for ad platform management, and perform network requests via curl as shown in references/generative-tools.md.
Sanitization: The skill does not mention any sanitization, filtering, or validation steps for the external performance data before processing it.

ad-creative