The Agent Skills Directory

[SAFE]: The skill incorporates extensive safety mechanisms for all fund-moving operations. This includes a mandatory dry-run mode by default, the requirement for a specific human-typed confirmation phrase derived from the rebalance plan, and compulsory transaction simulation via RPC before any submission. These protocols ensure the user maintains full control and oversight of their assets.
[EXTERNAL_DOWNLOADS]: The skill retrieves real-time market data, pool statistics, and strategy configurations from the official APIs of prominent Solana DeFi protocols, including Orca, Raydium, Meteora, and Kamino Finance. It also utilizes well-known data services like Birdeye and DefiLlama for historical price information and yield sanity checks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests market data and protocol statistics from external decentralized exchange (DEX) APIs to drive backtesting and yield comparison features.
Ingestion points: External data is fetched in templates/yield-compare.ts (DEX REST APIs), templates/kamino-vault.ts (Kamino API), and references/backtesting.md (Birdeye/DefiLlama OHLCV data).
Boundary markers: The skill processes structured JSON data from these established APIs, mapping numeric fields to internal logic, which significantly limits the risk of natural language instruction injection.
Capability inventory: The skill has the capability to generate transaction instructions for swaps and liquidity management, which are delegated to a core skill for final assembly and submission.
Sanitization: Execution is gated by human confirmation and strict validation of safety caps (e.g., maximum slippage and notional limits) after simulating the transaction outcome.
[COMMAND_EXECUTION]: The rebalancing and vault management components generate transaction instructions for Solana program interactions. These operations are managed via a simulate-first pipeline and are delegated to a core development skill (../solana-dev) for final signing and execution.

position-manager