The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and install new agent logic from arbitrary remote locations using the npx skills add command. This pattern is suggested whenever a skill.md file is encountered on a documentation site, allowing unverified external code to be integrated into the agent's operating context.
[COMMAND_EXECUTION]: The skill utilizes shell-based tools including curl and grep to fetch and search documentation. These tools are used on data obtained from untrusted external URLs, which could lead to command argument manipulation if the remote site serves malicious content.
[EXTERNAL_DOWNLOADS]: The skill frequently fetches content from external documentation sites to function. While this is the primary purpose, these downloads originate from untrusted sources provided at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests large amounts of potentially attacker-controlled content from documentation sites without sufficient safeguards.
Ingestion points: The skill fetches content from llms.txt, llms-full.txt, and markdown URL variants from external doc sites as described in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters mentioned to isolate the external content or warn the agent against embedded instructions.
Capability inventory: The skill has access to shell execution via curl, grep, and the npx package manager to add new skills.
Sanitization: No validation, escaping, or sanitization processes are described for the data retrieved from external sources before it is processed by the agent.

doc-reader