The Agent Skills Directory

[SAFE]: The skill provides legitimate documentation, code examples, and templates for integrating the DFlow protocol. All code follows standard Solana development practices, using environment variables for sensitive keys and providing clear transaction signing logic.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines an interaction model where the agent ingests market metadata (titles, event descriptions) from DFlow's APIs and has the capability to execute on-chain transactions. This constitutes a surface for indirect prompt injection, as malicious metadata could theoretically be used to influence an agent's trading behavior. This is an inherent risk for financial agent skills and is not indicative of malicious intent within the skill itself.\n
Ingestion points: Market and event metadata retrieved from https://api.prod.dflow.net (referenced in SKILL.md and examples/prediction-markets/query-markets.ts).\n
Boundary markers: None explicitly defined in the provided code examples to separate external data from system instructions.\n
Capability inventory: Transaction execution via connection.sendTransaction (found in examples/imperative-swaps/basic-swap.ts, examples/declarative-swaps/intent-swap.ts, and others) and network requests via fetch.\n
Sanitization: No specialized sanitization for LLM-mediated injection is present in the provided templates.\n- [EXTERNAL_DOWNLOADS]: The documentation guides users to install official DFlow protocol libraries from NPM (@dflow/solana-agent-kit, @dflow/clearpools) and references the project's official GitHub repositories (DFlowProtocol). These are standard, legitimate resources for developers integrating the protocol.

dflow