The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bundled Python scripts (scripts/fetch_page.py and scripts/build_timeline.py) to perform date arithmetic and retrieve live deadlines. These scripts are deterministic, use the standard library only, and do not employ unsafe patterns such as eval() or os.system() with unsanitized input.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch content from external URLs (conference websites) via fetch_page.py to ensure submission dates are accurate. The fetcher script implements robust safety and politeness features, including a 24-hour cache, rate limiting by host (persisted to .cache/plan-submission/), and a descriptive User-Agent that identifies the tool via a required CONTACT_EMAIL environment variable.
[DATA_EXFILTRATION]: While the skill performs network requests, these are scoped to fetching conference details. The CONTACT_EMAIL environment variable is used purely for identification in a polite User-Agent header, following established web scraping standards rather than representing a malicious exfiltration vector.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests and processes text from external conference websites.
Ingestion points: HTML content from conference pages is fetched and converted to text by scripts/fetch_page.py (relative path: scripts/fetch_page.py).
Boundary markers: The instructions do not explicitly mandate the use of delimiters when presenting the external content to the agent, which is a common vulnerability surface for untrusted text.
Capability inventory: The skill has the capability to execute local scripts and perform network reads.
Sanitization: The TextExtractor class in fetch_page.py effectively strips potentially dangerous HTML tags (<script>, <style>, etc.), which significantly reduces the risk of code-based injection, although the agent remains responsible for evaluating the resulting natural language text.

plan-submission