The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documents the use of highly sensitive credentials, including raw private keys and mnemonics, via command-line flags like --private-key and --mnemonic. For AI agents, this is a significant security risk as command-line arguments are frequently captured in logs, audit trails, and process listing tools, leading to potential credential theft.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the alkahest-cli global package via NPM and references the @alkahest/ts-sdk library. These are official vendor-provided tools required for the protocol's core functionality.
[COMMAND_EXECUTION]: The skill's operation relies on executing shell commands through the alkahest CLI utility. The agent is directed to perform complex blockchain operations, including token transfers and contract interactions, which involve high-privilege execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes arbitrary strings from external sources (e.g., fulfillment items in a StringObligation or blockchain event logs).
Ingestion points: Fulfillment data and attestation details retrieved via the alkahest CLI or SDK.
Boundary markers: No clear delimiters are defined to separate untrusted data from the agent's instructions.
Capability inventory: The agent can initiate token transactions, execute smart contract logic, and submit oracle decisions on-chain.
Sanitization: There is no evidence of sanitization or validation of external content retrieved from the blockchain before it is used to drive agent behavior.

alkahest-user