The Agent Skills Directory

[SAFE]: The skill follows a principle of least privilege by delegating all sensitive operations, such as price quoting, token swapping, and transaction signing, to external integration modules ('swap-integration' and 'viem-integration'). This design ensures the bot itself does not handle private keys or complex execution logic.
[SAFE]: Explicit security guardrails are embedded in the instructions, requiring the agent to perform strict regular expression validation on token addresses and numeric amounts. Furthermore, the skill mandates the rejection of any input containing shell metacharacters, effectively mitigating command injection risks.
[SAFE]: The skill follows secure credential management practices by instructing the agent to source API keys from environment variables rather than hardcoding them within the skill or configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, including user-provided configuration and persistent state files, which are interpolated into subsequent tool calls.
Ingestion points: Configuration values such as token addresses, buy amounts, and cadences are read from user input, and historical data is retrieved from a JSON state file (SKILL.md).
Boundary markers: Explicit delimiters for prompt interpolation are not defined, though state is managed in a structured format.
Capability inventory: The skill utilizes file system access for state management and interacts with trading and wallet integrations to execute financial transactions.
Sanitization: The skill provides strong sanitization instructions, including mandatory regex validation for addresses and amounts, and a comprehensive rejection list for shell metacharacters.

dca-bot