The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external news content from RSS feeds and web pages to generate summaries and reports, creating a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters via references/methods/rss.py and various web-scraping methods defined in the methods/ directory (e.g., using webfetch-smart or browser-smart).
Boundary markers: No explicit boundary markers or instructions to disregard embedded commands are present in the references/prompts/summary.md or references/prompts/report.md templates.
Capability inventory: The skill has the ability to write to the local filesystem (HTML/Markdown generation in build.py), modify a local SQLite database (scripts/db.py), and execute shell commands for git operations and deployment.
Sanitization: There is no evidence of sanitization or filtering of external content before it is interpolated into prompts for the LLM.
[EXTERNAL_DOWNLOADS]: The skill recommends the global installation of an unverified third-party tool, agent-browser, via NPM (npm install -g agent-browser). It also depends on several standard Python libraries for web scraping and feed parsing.
[COMMAND_EXECUTION]: The workflow relies on the execution of multiple shell commands, including database initialization (python3 scripts/db.py init), repository management via the GitHub CLI (gh repo create), and automated site building and deployment via Git.
[CREDENTIALS_UNSAFE]: The LOGIN_GUIDE.md file provides detailed instructions for users to provision credentials for Twitter/X. This includes manually extracting session cookies and setting plaintext usernames and passwords as environment variables (e.g., export TWITTER_PASSWORD="your_password"), which is a risky security practice.

daily-news