The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's example workflow uses npx to download and execute the package agent-skills-setup-for-antigravity. This package is not from a known trusted vendor or well-known service, posing a significant supply chain risk where arbitrary code could be executed in the environment.
Evidence: run: npx agent-skills-setup-for-antigravity run-skill daily-news-report in SKILL.md.
[COMMAND_EXECUTION]: The skill uses npx to run a setup and execution tool within a GitHub Actions environment. Executing unknown CLI tools allows for arbitrary command execution on the runner.
Evidence: run: npx agent-skills-setup-for-antigravity run-skill daily-news-report in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources (RSS feeds, Reddit, web scraping) and passes them directly to an LLM for summarization. The instructions lack boundary markers or sanitization to prevent the AI from following malicious commands embedded in the news articles.
Ingestion points: External RSS feeds, NewsAPI, Reddit API, and direct web scrapes via firecrawl-scraper (documented in SKILL.md).
Boundary markers: None identified. The prompt strategy in SKILL.md directly interpolates text without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill uses an LLM to summarize and classify content, then generates reports in Markdown, HTML, or Slack/Discord payloads (documented in SKILL.md).
Sanitization: No evidence of input sanitization, filtering, or validation of the ingested content before it reaches the LLM context.

daily-news-report