The Agent Skills Directory

[PROMPT_INJECTION]: The references/workflows/prompt-assembly.md file contains explicit instructions to bypass AI safety refusals regarding sensitive or copyrighted figures, directing the agent to create 'stylistically similar alternatives' and stating 'DO NOT refuse to generate'.
[EXTERNAL_DOWNLOADS]: Documentation in references/workflows/prompt-assembly.md suggests using npx to execute a script (scripts/main.ts) that is not included in the skill's source files, which introduces a dependency on unverified external code.
[COMMAND_EXECUTION]: The skill instructs the agent to use bash to probe the filesystem for configuration files, including paths within the user's home directory ($HOME). While these are associated with the vendor's naming conventions, the use of shell commands for filesystem probing is a significant capability.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content and interpolates it into prompts for image generation tools without explicit sanitization or instructions to ignore embedded commands. Ingestion points: User text pastes and file inputs during Step 1. Boundary markers: The prompt template uses horizontal rules (---) and markdown headers. Capability inventory: Shell command execution, filesystem write access, and external tool usage. Sanitization: No sanitization or filtering is defined for the input content.

baoyu-xhs-images