The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from external web sources through the AnyCap search and crawl functionalities. This creates a vulnerability surface where adversarial content hosted on the web (e.g., in Wikipedia articles or general search results) could contain hidden instructions designed to influence the agent's behavior.
Ingestion points: Data enters the agent context through anycap search results in references/lookup.md, references/stats.md, and references/history.md, as well as via anycap crawl of external URLs in references/verify.md.
Boundary markers: The skill employs structural boundaries by requesting JSON-formatted output and using jq for parsing, which helps separate data from instructions. It also uses sed to strip markdown fences.
Capability inventory: The agent is granted capabilities to execute shell commands (anycap, jq, sed, column) and perform local file writes (e.g., to squad.json or /tmp/player.md).
Sanitization: The instructions focus on formatting and data extraction but do not explicitly detail sanitization or filtering of the retrieved web content to prevent the execution of embedded instructions.
[COMMAND_EXECUTION]: The skill's core workflow relies on the agent generating and executing shell commands that include external parameters such as player names, team names, and user-defined dimensions. While this is necessary for the CLI-based functionality, it requires the agent to handle these inputs securely to prevent potential command injection or shell metacharacter abuse.

anycap-worldcup-predict