The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided Markdown files and site-specific configuration files such as writing-style.md and AGENTS.md. These files are parsed to extract rules and content for validation. An attacker could potentially embed malicious instructions within these files to influence the agent's behavior during the validation process.
Ingestion points: Target Markdown file (e.g., index.md), hugo/AGENTS.md, and content/prompts/writing-style.md.
Boundary markers: No explicit delimiters or safety instructions are provided to the agent to prevent it from obeying instructions embedded within the processed files.
Capability inventory: The skill can execute shell commands (bash, python3, hugo, lychee, markdownlint-cli2) and invoke other agent skills (ai-sanitize, write-check-links, write-run-markdown-lint).
Sanitization: There is no evidence of sanitization or escaping of the external content before it is processed or passed to the validation scripts.
[DATA_EXFILTRATION]: The skill performs automated network requests to external domains using the lychee link-checking tool. These requests target URLs found within the user-provided Markdown content to verify their validity. While this is a primary function of the skill, it involves interaction with non-whitelisted external domains.
[COMMAND_EXECUTION]: The skill performs its core validation tasks by executing shell commands. This includes running a Hugo site build to check for errors, executing Python scripts for front-matter and phrase scanning, and running CLI tools for link checking and linting.
[DYNAMIC_EXECUTION]: The skill dynamically computes paths to its utility scripts using environment variables and relative directory structures (e.g., referencing a sibling hugo-agents-compliance directory). It then executes scripts from these resolved paths. While these scripts are associated with the same vendor, this pattern of dynamic execution is a notable security surface.

write-validate-article