The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The SKILL.md file contains instructions for the agent to fetch WordPress categories and tags using curl -u "$WORDPRESS_USERNAME:$WORDPRESS_APP_PASSWORD". On multi-user systems, this practice exposes sensitive credentials in the process list, which can be accessed by other users or processes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and interpret files from the user's project to understand feature implementations for writing blog posts.
Ingestion points: Files read in the project directory during Phase 1.3 of SKILL.md and git logs.
Boundary markers: The instructions lack delimiters or specific directives to prevent the agent from executing instructions potentially hidden within the analyzed codebase.
Capability inventory: The skill possesses file system access and network capabilities (via curl and a Python script).
Sanitization: There is no filtering or validation mechanism defined for the content read from the codebase before it is passed to the LLM.

technical-writing