The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation provides shell commands to download and install multiple supplementary markdown files (e.g., AUTH.md, WALLETS.md, FEES.md) from the bags.fm domain, introducing a supply chain risk through unverified remote content.- [CREDENTIALS_UNSAFE]: The API defines a /agent/wallet/export endpoint for retrieving wallet private keys. Accessing and handling raw private keys within an AI agent environment is a high-risk security practice.- [DATA_EXFILTRATION]: The ability to export private keys via an API call, combined with the agent's general network capabilities, creates a direct path for the exfiltration of sensitive cryptographic material.- [COMMAND_EXECUTION]: The installation and setup guides include bash snippets that perform filesystem operations and download remote files directly into user-specified directories on the local machine.- [CREDENTIALS_UNSAFE]: The skill documentation instructs the agent to store sensitive authentication materials, including long-lived JWT tokens and API keys, in a persistent local configuration file (~/.config/bags/credentials.json).- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. 1. Ingestion points: Untrusted data is ingested from public-api-v2.bags.fm, specifically token metadata and fee share configurations. 2. Boundary markers: No delimiters are specified to isolate this external data. 3. Capability inventory: The skill can list wallets, export private keys, and execute Solana transactions. 4. Sanitization: No sanitization or validation of the ingested API data is performed.

bags