The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the 'filmcraft-cli' binary and application DMGs from GitHub releases (github.com/storytold and github.com/full-aigc-skills). All downloads are validated against SHA-256 hashes pinned in the skill's runtime.lock.json and desktop.lock.json files.
[REMOTE_CODE_EXECUTION]: The primary purpose of the skill is to download, verify, and execute specialized media editing binaries. This execution is gated by a robust internal permission system and hash verification to ensure only authentic vendor code is run.
[COMMAND_EXECUTION]: The skill makes extensive use of the Python subprocess module to manage the lifecycle of the media tool and its helpers. These executions are carefully wrapped in security filters.
[INDIRECT_PROMPT_INJECTION]: The skill accepts JSON-formatted editing plans which represent an ingestion surface for indirect prompt injection. This risk is mitigated through a command registry that validates all requested command IDs and strictly enforces parameter types, such as 64-bit integer 'tick' validation for timeline operations.
[SAFE]: The skill implements a sophisticated security boundary using macOS sandbox-exec. This sandbox enforces a 'deny-by-default' policy, specifically blocking network access (except for explicit maintenance) and restricting file system access to paths explicitly authorized by the user via command-line arguments. Additionally, the ZIP extraction logic in bootstrap.py is hardened against path traversal (ZipSlip) and symbolic link attacks.

filmcraft-cli-transcript