The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to read local project configuration files (.claude/project-context.md or .cursor/project-context.md) to retrieve keywords and content strategies. This creates a surface where malicious instructions embedded in those project files could influence the agent's behavior during the SEO optimization process.
Ingestion points: The agent reads data from .claude/project-context.md and .cursor/project-context.md as specified in the 'Initial Assessment' section of SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between valid project data and potentially malicious embedded instructions in these external files.
Capability inventory: The skill uses standard file-reading capabilities and generates text-based output (titles, descriptions, tags) based on the ingested data.
Sanitization: No sanitization, validation, or filtering of the content retrieved from the project context files is mentioned or performed.

youtube-seo