effectcraft-social-loop

Warn

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references a specific URL (https://github.com/storytold/effectcraft/releases/download/v0.10.0/effectcraft-0.10.0-windows-x64-portable.zip) in references/craft-command-evidence.json to download a portable Windows binary. While a SHA256 hash is provided for integrity, the source repository is an external third-party that is not recognized as a well-known or trusted service.
  • [REMOTE_CODE_EXECUTION]: The skill provides instructions and metadata for downloading and running a remote binary (effectcraft-cli.exe). This involves executing code from a non-whitelisted external source within the agent's environment.
  • [COMMAND_EXECUTION]: In references/craft-methods.md, the skill uses the command_run tool to execute effectcraft-cli.exe for creating video compositions, adding shape layers, and setting property keyframes. This involves running an externally downloaded executable with parameters derived from the task context.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided media assets and project files, creating a surface for indirect instructions.
  • Ingestion points: SKILL.md (task starting point reads materials) and references/craft-methods.md (reading input and existing project files like input.ecproj).
  • Boundary markers: None found.
  • Capability inventory: SKILL.md and references/craft-methods.md detail tools like command_run, save_project, and save_file.
  • Sanitization: None found.
  • [PERSISTENCE]: SKILL.md suggests installing an additional skill (effectcraft-editor) via npx skills add full-aigc-skills/effectcraft-skills, which is a persistent environment modification using the author's own resources.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 23, 2026, 09:37 AM