The Agent Skills Directory

[METADATA_POISONING]: There are inconsistencies between the skill's frontmatter and its accompanying files. The frontmatter identifies the author as 'ck' and the license as 'MIT', whereas the LICENSE file specifies 'Injective Labs' as the copyright holder and the 'Apache 2.0' license. Such discrepancies can mislead users or automated systems regarding the skill's origins and legal terms.
[CREDENTIALS_UNSAFE]: The generate_wallets_from_seed function in SKILL.md explicitly extracts and returns the private_key of generated wallets in plaintext hex format. While required for wallet derivation tasks, storing or returning raw private keys in the agent's context or conversation history increases the risk of credential exposure if logs are compromised or if the agent inadvertently reveals these values to the user.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through the processing of untrusted external data.
Ingestion points: The skill accepts a seed_phrase in the generate_wallets_from_seed function and a raw address string in the normalize function.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the processed inputs.
Capability inventory: The skill possesses significant capabilities, including transaction signing and broadcasting via MsgBroadcasterWithPk, and querying account balances through network RPC calls.
Sanitization: While the skill provides regex validation for addresses (INJ_BECH32, ETH_HEX), it lacks validation or sanitization for the seed_phrase input.

injective-wallet-ops