The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation includes hardcoded Ethereum private keys and mnemonics which are sensitive cryptographic secrets.
Evidence: references/accounts-and-keys.md contains a specific private key (0xac0974bec39a17e36ba4a6b4d238ff944bacb478cbed5efcae784d7bf4f2ff80) and a mnemonic phrase (legal winner thank year wave sausage worth useful legal winner thank yellow).
Although these are labeled as test credentials for the Anvil development environment, hardcoding real cryptographic secrets in skill instructions is a risky practice that could lead to accidental use in production environments.
[METADATA_POISONING]: There is a discrepancy between the declared author in the skill's metadata and the actual provider of the skill, which is potentially deceptive.
Evidence: SKILL.md lists author: uniswap in its YAML frontmatter, whereas the skill is authored by blockmatic. This mismatch could mislead users regarding the origin, official support, or security vetting of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external blockchain sources, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The skill defines patterns for fetching untrusted data from the Ethereum blockchain using functions like client.getLogs(), client.readContract(), and client.getTransaction() in references/reading-data.md.
Boundary markers: The skill includes an "Input Validation Rules" section in SKILL.md which prescribes validation for Ethereum addresses and RPC protocols to mitigate basic injection attempts.
Capability inventory: The skill instructions enable the agent to execute transactions and write to smart contracts via client.sendTransaction() and client.writeContract() as documented in references/writing-transactions.md.
Sanitization: While the skill recommends using isAddress() and parseUnits(), the processing of arbitrary blockchain data (like contract return values or event logs) represents a trust boundary where malicious on-chain data could attempt to influence the agent's logic.

viem-v2