The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from pull request bodies, human and agent review comments, and spec files to generate interactive D3 visualizations. This creates an attack surface where malicious input in a PR or comment could attempt to influence the generated walkthrough or target the reviewer's browser.
Ingestion points: PR body, human/agent review comments, and PR-changed spec files (e.g., PRODUCT.md) gathered via gh and git commands.
Boundary markers: The instructions do not define specific prompt delimiters for the ingested data, although the data is structured as JSON before being passed to the visualization runtime.
Capability inventory: The skill writes a local HTML/JS walkthrough to .warp/pr-walkthrough/index.html, executes validation scripts using Python, and can deploy the result to Cloudflare Pages via npx wrangler.
Sanitization: The generated D3 runtime script (scripts/d3_canvas_runtime.py) implements an escapeHtml helper to sanitize node titles, summaries, details, comments, and links before rendering them in the UI.
[EXTERNAL_DOWNLOADS]: Fetches the D3.js library from jsdelivr.net. The skill strictly follows security best practices by requiring a pinned official release (7.9.0) rather than dynamic version ranges.
[EXTERNAL_DOWNLOADS]: Optionally uses npx wrangler for deploying walkthroughs to Cloudflare Pages. Cloudflare is a well-known service, and the instructions require user-initiated deployment.
[COMMAND_EXECUTION]: Utilizes git and gh (GitHub CLI) to perform repository analysis and metadata collection. These commands are standard for PR-related tooling and are executed within the local repository context.

pr-walkthrough