The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data from various external web sources (e.g., Hacker News, arXiv, The Verge, GitHub Trending) during its information collection phase. This creates a surface for indirect prompt injection where malicious instructions embedded in the fetched content could influence the agent's summary or report generation.
Ingestion points: Multiple web_fetch calls in SKILL.md (Step 2) retrieve content from third-party sites.
Boundary markers: The skill does not use specific delimiters or instructions to ignore potential commands within the fetched content.
Capability inventory: The skill can execute shell commands via scripts/md_converter.py to install packages and write files to the workspace.
Sanitization: The skill provides a degree of semantic sanitization by instructing the agent to summarize content based on "AI domain knowledge" rather than direct translation.
[REMOTE_CODE_EXECUTION]: The script scripts/md_converter.py automatically attempts to install several Python packages (weasyprint, pdfkit, xhtml2pdf, fpdf2) using pip install via subprocess.run if they are not already present on the system. These libraries are downloaded from the official Python Package Index (PyPI).
[COMMAND_EXECUTION]: The skill executes shell commands using subprocess.run within scripts/md_converter.py to manage dependencies and perform file format conversion tasks.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch AI-related information from various domains and to download required Python libraries from PyPI.

ai-daily-report