The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple bash script blocks to manage its environment. These include checking for updates via rmbc-update-check, logging analytics via rmbc-analytics, and setting up workspace contexts via rmbc-workspace. These scripts interact with files in ~/.rmbc-skills/ and the skill's own directory.
[COMMAND_EXECUTION]: The skill uses sed -i to programmatically update telemetry settings within the user's local config.yaml file based on user input.
[EXTERNAL_DOWNLOADS]: The skill provides an instruction to open an introductory video on YouTube, which is a well-known service.
[DATA_EXFILTRATION]: The skill logs usage data (skill name, product name, and a tier level) through a vendor-provided analytics script. The skill explicitly states that this data is anonymous and does not include code, prompts, or file paths.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted user data for ad generation.
Ingestion points: User-provided inputs including product_description, target_audience, and key_mechanism (found in SKILL.md).
Boundary markers: Absent.
Capability inventory: The skill has access to shell execution and filesystem operations.
Sanitization: No explicit sanitization or escaping of these inputs was observed before their use in generation logic or script arguments.

ad-angle-generator