The Agent Skills Directory

[COMMAND_EXECUTION]: The file AUDIT_PROMPT.md directs the agent to execute a series of shell commands, including grep, awk, find, and jq. These are used to parse project frontmatter and analyze bundle sizes in the dist/ directory.\n- [COMMAND_EXECUTION]: The skill instructs the agent to create a local bash script (audit-blog-post.sh) and use chmod +x to grant it execution permissions, which involves modifying file system privileges to execute unverified script logic.\n- [PROMPT_INJECTION]: In references/human-voice.md and references/validation.md, the agent is instructed to bypass AI detection mechanisms by utilizing 'Intentional Imperfections.' This includes purposely misspelling words like 'thier', 'recieve', and 'definately' to evade automated filters that distinguish between human and AI-generated content.\n- [REMOTE_CODE_EXECUTION]: The audit workflow requires the agent to install an external package (npm install -g lighthouse) and execute the CLI tool against a user-specified URL, which triggers remote data fetching and execution in a shell environment.\n- [EXTERNAL_DOWNLOADS]: The skill performs external downloads via the npm package manager to install the Lighthouse tool and manage project dependencies during the npm run build phase.

astro-blog