The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from X (Twitter) timelines to generate AI-powered semantic summaries and daily reports. This creates a vulnerability where an attacker could post specifically crafted tweets designed to influence the agent's behavior during the summarization or reporting process.
Ingestion points: scripts/subscribe.py fetches tweet text and metadata via the userTimeline API.
Boundary markers: No specific delimiters or instructions to ignore embedded commands were found in the prompt construction for the AI summary.
Capability inventory: The skill can write HTML files to the local disk, open these files in the default system browser via subprocess.run, and output Markdown tables for the agent to process.
Sanitization: While scripts/generate_report.py performs standard HTML escaping for the final report, the natural language content used for the AI semantic summary is not explicitly sanitized against prompt injection patterns.
[CREDENTIALS_UNSAFE]: The sample data file output/x_top_accounts.json contains several hardcoded AWS S3 pre-signed URLs which include AWS credentials (e.g., X-Amz-Credential=AKIAVSTJR7ZBRQ76TM7R). While these are likely restricted-scope keys used for public image hosting by the vendor, hardcoding credentials in distributed files is a security anti-pattern.
[EXTERNAL_DOWNLOADS]: The HTML daily report generated by scripts/generate_report.py dynamically loads the html2canvas and jspdf JavaScript libraries from cdn.jsdelivr.net. These are well-known, established services used for report generation features (exporting to images or PDF).
[COMMAND_EXECUTION]: The skill uses subprocess.run in two contexts: scripts/subscribe.py executes generate_report.py to process fetched data, and scripts/generate_report.py uses platform-specific commands (like open, start, or xdg-open) to automatically launch the generated HTML report in the user's browser.

x-subscribe