The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the project binary locally with flags such as --help to observe its behavior. This execution is limited to discovery for the purpose of contract reconstruction and is restricted to the local environment.- [DATA_EXFILTRATION]: No network activity was detected. The skill only interacts with the local filesystem at the user-specified project path and does not transmit data externally.- [PROMPT_INJECTION]: The skill instructions include explicit guardrails to prevent the AI from inventing behavior or bypassing the observed implementation. These measures, combined with mandatory user approval gates for all generated artifacts, provide strong protection against instruction-based attacks.- [CREDENTIALS_UNSAFE]: The skill accesses only standard project files (Cargo.toml, README.md, etc.) and does not seek access to sensitive credential stores or environment variables.- [REMOTE_CODE_EXECUTION]: All operations are performed locally on the existing project files. There are no instructions to download or execute code from external repositories or unknown third-party services.

cli-forge-takeover