The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary workflow involves the agent reading Python templates (e.g., templates/basic_fetch.py, templates/stealth_cloudflare.py), populating them with user-provided parameters like URLs and CSS selectors, and executing the resulting scripts using the Bash tool.
[EXTERNAL_DOWNLOADS]: The instructions require installing the scrapling Python package and its browser-based fetchers using commands such as pip install "scrapling[fetchers]" and scrapling install.
[PROMPT_INJECTION]: The skill is inherently exposed to indirect prompt injection due to its function of scraping and processing external web content.
Ingestion points: Untrusted HTML and text are retrieved from third-party websites via the generated scripts and printed to the agent's context (e.g., in templates/basic_fetch.py).
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat scraped data as untrusted or to ignore instructions contained within it.
Capability inventory: The agent has access to the Bash tool and file-writing capabilities across the entire skill folder, which could be targeted by malicious instructions embedded in scraped pages.
Sanitization: Scraped content is presented to the agent without any prior sanitization, filtering, or validation.
[DATA_EXFILTRATION]: The skill implements a persistence mechanism for session cookies, instructing the agent to save them to references/cookie-vault.md. While intended for legitimate session maintenance across scraping tasks, this creates a local store of sensitive credentials that could be exposed or exfiltrated if the agent processes malicious scraped content.

scrapling