The Agent Skills Directory

[OBFUSCATION]: The skill package includes eight undocumented reference files (e.g., references/architecture-patterns.md, references/deployment-pipelines.md, references/security-best-practices.md) that are not listed in the SKILL.md reference section. Each of these files contains 150 sections of repetitive, non-functional text discussing kernel-level optimizations (e.g., epoll_wait, io_uring) which are entirely irrelevant to the skill's stated purpose of web accessibility. This data flooding pattern is a common technique used to overwhelm security scanners and exhaust the AI agent's context window.
[COMMAND_EXECUTION]: The instructions and reference files contain multiple shell command templates for tool installation and CI/CD integration (e.g., npm ci, npx axe, pa11y-ci). While these tools are standard in the industry, the presence of pre-written shell snippets can be misused by an agent to execute commands without sufficient oversight.
[EXTERNAL_DOWNLOADS]: The skill documentation (e.g., SKILL.md, references/testing-tools.md) references several external dependencies and GitHub Actions, such as axe-core, Lighthouse, and treosh/lighthouse-ci-action. These are well-known services from reputable sources, but the skill facilitates the downloading and execution of this external code within the agent's workflow.
[METADATA_POISONING]: The skill uses misleading metadata by including a vast amount of

design-accessibility