The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted WeChat chat logs, creating an indirect prompt injection surface where malicious messages could influence the AI's summary or behavior.
Ingestion points: scripts/analyze_chat.py reads user-provided JSON chat logs and extracts message content.
Boundary markers: No explicit delimiters or instructions are used in references/ai_prompt.md to isolate untrusted chat data from the AI's core instructions.
Capability inventory: The AI's output is written to ai_content.json and subsequently rendered into an HTML/PNG report using scripts/generate_report.py.
Sanitization: The analyze_chat.py script performs basic text cleaning (removing newlines and truncating) but does not sanitize or escape content against injection attacks.
[COMMAND_EXECUTION]: scripts/generate_report.py executes the Playwright library to launch a headless Chromium browser instance. While this is necessary for the skill's primary function of generating images, it represents the execution of a high-privilege subprocess.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the Chromium browser binary via the playwright install chromium command as part of its setup process.
[DYNAMIC_EXECUTION]: The report generation logic in scripts/generate_report.py uses the Jinja2 templating engine to insert chat data into assets/report_template.html. The Jinja2 environment is initialized without autoescape=True, which means any HTML or JavaScript contained within the chat logs will be rendered directly. This content is then executed by the Playwright browser instance, potentially leading to local script execution (XSS) during the image rendering process.

wechat-daily-report