The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from DefiLlama's API endpoints, which includes information like protocol names, descriptions, and slugs. Because this data is often sourced from third-party protocol developers, it could contain malicious instructions designed to influence the agent's behavior when the data is rendered or processed.
Ingestion points: API calls to api.llama.fi, pro-api.llama.fi, and bridges.llama.fi in exports.py and the usage templates in SKILL.md.
Boundary markers: The skill does not use specific delimiters or instructions to prevent the agent from obeying instructions embedded within the API response data.
Capability inventory: The skill uses network operations (proxied_get) and expects the agent to execute shell commands (bash) to run its Python exports.
Sanitization: There is no explicit sanitization of string fields returned by the API before they are processed or presented to the agent context.
[COMMAND_EXECUTION]: The skill is designed as a 'script-mode' PoC, requiring the AI agent to generate and execute Python code via a bash shell to access its core functions located in exports.py.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data from well-known DeFi analytics services, including api.llama.fi, pro-api.llama.fi, bridges.llama.fi, and stablecoins.llama.fi.

defillama