The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads sensitive files such as .env.local to verify API keys and src/lib/config/tenant.config.ts to gather internal brand context. Accessing environment variables and configuration files can lead to the exposure of internal project secrets and architecture details.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data from $ARGUMENTS and interpolates it directly into generative AI prompts without sanitization, escaping, or boundary markers. This creates a surface where malicious instructions in asset descriptions could influence the agent's behavior. Ingestion points: $ARGUMENTS variable in SKILL.md. Boundary markers: Absent around the interpolated description variable. Capability inventory: The agent has Write and Bash capabilities and is instructed to modify the local file system. Sanitization: None.
[DYNAMIC_EXECUTION]: The skill includes TypeScript code templates that the agent is expected to execute to interact with the Google Generative AI SDK and perform file system writes using the fs module.

asset-series