The Agent Skills Directory

Data Exposure & Credentials (MEDIUM): The skill requires a path to an Arweave JWK wallet file (wallet_path) in its configuration. Accessing private keys is a high-risk operation. While necessary for the skill's primary purpose of signing blockchain transactions, users must ensure the wallet file is properly secured and contains only necessary funds.
External Downloads & Unverifiable Dependencies (MEDIUM): The installation instructions require cloning from github.com/lobster-skills/pasp, which is not a trusted source according to the [TRUST-SCOPE-RULE]. The skill also depends on several external NPM packages that are installed at runtime.
Indirect Prompt Injection (LOW): The skill ingests untrusted data from the permaweb via queryPosts, getProfile, and getThread commands.
Ingestion points: Data is fetched from Arweave GraphQL and transaction data fields.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present when processing fetched content.
Capability inventory: The skill has arweave.upload capabilities, allowing it to write data back to the blockchain.
Sanitization: Uses the marked library to render Markdown to HTML, providing basic structural parsing but no protection against prompt injection within the text itself.
Data Exfiltration (LOW): The skill uploads data to arweave.net. While this is the intended functionality of a permaweb social protocol, it involves sending data to a non-whitelisted domain. Given it also accesses a wallet file, there is a theoretical risk of key exfiltration if the code were malicious, though no such logic was detected in the provided source.

pasp