The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external website content and local source code, which are untrusted data sources. This constitutes an indirect prompt injection attack surface. However, the risk is mitigated as the skill's output is structured (JSON/Markdown) and the provided dashboard generator utilizes JSON.stringify for data serialization, preventing cross-site scripting (XSS) or prompt leakage in the final artifacts. The ingestion is necessary for the skill's primary purpose.
Ingestion points: Live URL fetching via WebFetch and local file scanning via Read and Glob (documented in rules/seo-workflow.md).
Boundary markers: None identified.
Capability inventory: File read/write access to the .hyper/seo-maker/ directory, execution of the bundled scripts/render-dashboard.mjs, and network access for auditing.
Sanitization: The dashboard rendering logic in scripts/render-dashboard.mjs properly serializes audit data into a JavaScript variable using JSON.stringify for safe handling in the browser.
[COMMAND_EXECUTION]: The skill requires the execution of its bundled script scripts/render-dashboard.mjs using bun. This is an internal operational step for generating the dashboard artifact and is consistent with the skill's design. The script itself performs atomic file operations and utilizes standard Node.js/Bun modules safely.
[EXTERNAL_DOWNLOADS]: The generated dashboard.html template references the Chart.js library from a well-known and trusted Content Delivery Network (JSDelivr). This is a standard practice for web-based data visualization and follows safety guidelines for well-known services.
[PRIVILEGE_ESCALATION]: The rendering script uses fchmodSync to set permissions on generated artifacts. This is used exclusively to ensure that updated files maintain the same file modes as the existing snapshots they replace, which is a standard maintenance operation and does not represent an escalation attempt.

seo-maker