The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads binary artifacts (ZIP and DMG files) from GitHub repositories associated with the author (full-aigc-skills) and the upstream project (storytold/filmcraft). These downloads are restricted to specific whitelisted URL prefixes and are subject to strict SHA256 hash verification against pinned values in the skill's lock files before any extraction or execution occurs.
[REMOTE_CODE_EXECUTION]: The skill downloads and executes a native CLI tool and a desktop application. While this pattern typically carries risk, it is the primary purpose of the skill. The risks are mitigated by verifying the integrity of all binaries using hardcoded SHA256 hashes and by executing the tools within a strictly defined macOS sandbox environment.
[COMMAND_EXECUTION]: The skill makes extensive use of the Python subprocess module to orchestrate the native CLI and media processing tools. All such executions are wrapped in a macOS 'sandbox-exec' profile generated by the skill's permission module, which implements a 'deny default' policy, restricts network access, and limits filesystem operations to specific directories explicitly authorized by the user via command-line arguments.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection as it processes JSON-based 'command plans' and external media metadata which could theoretically contain malicious instructions. This risk is effectively mitigated by the skill's design, which uses structured data validation, parameter whitelisting for native commands, and a robust execution sandbox that prevents unauthorized side effects even if the underlying tool were influenced by malicious input.

filmcraft-cli-color