The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install an unverified third-party Python package blockrun-llm via pip. This dependency is not associated with a recognized trusted organization and acts as a gateway for external network operations.
[CREDENTIALS_UNSAFE]: The skill manages a sensitive wallet session file located at $HOME/.blockrun/.session. This file contains private financial credentials (cryptocurrency wallet state) that are stored locally, creating a target for data exposure or exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves content from external sources such as X (Twitter) and various third-party LLMs, which is then processed by the agent without sufficient safety boundaries.
Ingestion points: Data enters the agent context through the client.chat() and client.generate() methods used in SKILL.md to fetch search results and model responses.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are provided to protect the agent from malicious instructions within the external data.
Capability inventory: The skill utilizes network access (via the SDK) and local file system writes (for wallet storage) across its instructional examples.
Sanitization: There is no evidence of sanitization or filtering of the untrusted data retrieved from external providers before it is presented to the agent.

blockrun