The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary user-provided content and explicitly requires the agent to preserve this data verbatim when constructing prompts for an image generation skill. This creates a surface where malicious instructions embedded in the source text could influence the behavior of the downstream process.
Ingestion points: User content is ingested from files or text input and saved to source.md in Step 1.2.
Boundary markers: The references/base-prompt.md template interpolates content using {{CONTENT}} without delimiters (like XML tags) or instructions to treat the data as untrusted.
Capability inventory: The skill invokes an image generation skill in Step 6, which uses the instructions generated from the user's input.
Sanitization: The skill's core principles and the references/analysis-framework.md mandate verbatim extraction and forbid summarization or rephrasing, which ensures any injection payloads remain intact.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to locate configuration files.
Evidence: Step 1.1 in SKILL.md uses test -f to search for EXTEND.md in both the project directory and the user's home directory ($HOME/.baoyu-skills/). While these are standard configuration checks, they involve automated command execution targeting the user's file system.

baoyu-infographic