The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scanners have detected a malicious URL in the file references/tokens/xls-0024.md. The URL https://x-tokenize.mypinata.cloud/ipfs/QmNpi8rcXEkohca8iXu7zysKKSJYqCvBJn3xJwga8jXqWU is identified as a phishing risk.
[EXTERNAL_DOWNLOADS]: The skill includes utility scripts (scripts/list-xls.sh and scripts/fetch-xls.sh) that use curl to download content from api.github.com and raw.githubusercontent.com. These operations involve fetching external data that is subsequently processed and read by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture creates a surface for indirect prompt injection. 1. Ingestion points: The fetch-xls.sh script downloads markdown specification files from a remote repository and introduces them into the agent's context. 2. Boundary markers: Absent. The skill does not use delimiters or instructions provided to the agent to treat the fetched content as untrusted or to ignore embedded instructions. 3. Capability inventory: The agent can execute the included bash scripts, read files, and perform network requests. 4. Sanitization: The scripts/extract-spec.py script used for processing retrieved content does not sanitize for injection patterns. Notably, it is explicitly configured to remove sections labeled 'security consideration', 'security', 'trust assumption', and 'data privacy', which may prevent the agent from being aware of critical security constraints while implementing the protocols.
[COMMAND_EXECUTION]: The skill provides and encourages the use of shell scripts (list-xls.sh, fetch-xls.sh) that perform network operations and directory searches, suggesting the agent run these commands to fulfill requests.

xrpl-standards