The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted HTML data, creating a surface for indirect prompt injection attacks where the converted Markdown might contain instructions that mislead the AI agent.
Ingestion points: Untrusted data enters the agent context through HTML files, standard input (stdin), or variables in the Python/Rust code snippets.
Boundary markers: The instructions lack explicit requirements for the agent to use delimiters (like code blocks) or specific 'ignore' instructions when outputting the converted content to prevent the agent from following instructions embedded in the converted text.
Capability inventory: The skill uses a CLI tool and libraries to transform data, which could then be passed to other agent tools or interpreted as instructions.
Sanitization: While the skill documents the --preprocess and --strip-tags flags to filter script and style tags, these do not prevent natural language prompt injections embedded in standard text nodes.
[COMMAND_EXECUTION]: The skill instructs the agent on how to execute the html-to-markdown CLI tool.
Evidence: Multiple examples provided in SKILL.md demonstrate the use of shell commands to process local files and pipe stdin to the utility.

converting-html