The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes documentation and example code for an AI-powered web scraper that processes untrusted external data and passes it to an LLM.
Ingestion points: The example in references/browser-rendering.md uses Puppeteer to fetch page content (page.content()).
Boundary markers: The code snippet does not implement boundary markers or instructions to ignore embedded prompts in the scraped data.
Capability inventory: The skill includes scripts capable of system command execution (scripts/cloudflare_deploy.py) and file system access.
Sanitization: No sanitization or validation of the scraped web content is present before it is interpolated into the LLM prompt.
[COMMAND_EXECUTION]: The deployment utility scripts/cloudflare_deploy.py uses subprocess.run to execute Cloudflare Wrangler commands.
Evidence: The script builds commands such as wrangler deploy by concatenating strings derived from configuration files (wrangler.toml) and command-line arguments.
Context: While the script uses argparse with defined choices for environment names to limit input, it inherently relies on shell execution for its primary functionality.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for installing official platform tools from external sources.
Evidence: references/gcloud-platform.md contains commands to download and execute the Google Cloud SDK installer (curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-linux-x86_64.tar.gz | bash).
Context: These downloads target established domains (google.com) and are standard procedures for setting up the Google Cloud environment.

devops