The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the gmgn-cli binary using subprocess.run in dev_score.py to fetch blockchain data.
Arguments are passed as a list, which prevents shell-level command injection.
The execution environment is configured via the ENV dictionary to handle rate-limiting parameters for the CLI.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted external data (token names and symbols) which are attacker-controlled and could contain malicious instructions designed to influence the agent or UI.
Ingestion points: External data is fetched from blockchain APIs via gmgn-cli calls in dev_score.py (e.g., portfolio activity).
Boundary markers: The script uses 「」 brackets to wrap strings identified as non-standard tickers, helping the agent distinguish data from its own instructions.
Capability inventory: The skill has the capability to execute the gmgn-cli binary and read its output.
Sanitization: The safe() function in dev_score.py cleanses input by removing non-printable characters, stripping ANSI escape sequences, and capping string length to prevent UI/report forgery or terminal manipulation.

gmgn-dev-score