The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from repository files and git diffs to generate an HTML report. This content is embedded into the report template without explicit instructions for sanitization or escaping. If the code being explained contains malicious HTML or script tags, they could be executed in the user's browser context when the report is opened.
Ingestion points: Content originates from git diff and file reads across the current branch.
Boundary markers: The skill lacks instructions to sanitize or wrap untrusted code snippets to prevent script execution within the HTML output.
Capability inventory: The skill uses git for data retrieval, mkdir for path preparation, and open to trigger browser execution of the generated file.
Sanitization: No sanitization logic is requested; the agent is instructed to write code directly into HTML <code> elements.
[EXTERNAL_DOWNLOADS]: The report template (references/page-template.html) fetches external resources for syntax highlighting and diagram rendering.
Evidence: Links to cdnjs.cloudflare.com for highlight.js and cdn.jsdelivr.net for Mermaid.js. These are well-known and established CDNs.
[COMMAND_EXECUTION]: The skill executes several shell commands to gather context and display the result.
Evidence: Uses git merge-base, git diff, git remote get-url, and gh pr view to extract repository metadata. It also uses open to launch the generated report in the default web browser.

explain-changes