The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to be executed via a shell command (node scripts/suggest.mjs) to generate a markdown report from local data directories. This is its intended primary purpose.
[DYNAMIC_EXECUTION]: The main script dynamically loads check modules from the local scripts/checks/ directory using import(). This is a standard plugin-style architecture for a linter or auditing tool and is restricted to local files provided by the skill itself.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted HTML snapshots and metadata from the site being analyzed. It parses this content to identify flaws (e.g., placeholder text, missing headers). While it processes untrusted data, it only extracts snippets for reporting purposes and does not execute any instructions contained within that data.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill reads local files from user-specified directories (capture/, baseline/, review/). However, it contains no network functionality and explicitly states it never requests the live site, mitigating the risk of data exfiltration.
[SAFE_PRACTICES]: The skill demonstrates safe coding practices, such as enforcing a minimum Node.js version and performing read-only operations on the provided input data.

suggest-improvements