anicraft
Fail
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPERSISTENCEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides a shell script that downloads a binary from an untrusted external repository and executes it.\n
- File: SKILL.md\n
- Evidence: Script downloads anicraft-cli from github.com/storytold/anicraft and runs it with the version flag.\n- [EXTERNAL_DOWNLOADS]: The skill fetches binary assets from a repository that is not a known or trusted vendor.\n
- File: SKILL.md\n
- Evidence: https://github.com/storytold/anicraft/releases/download/v0.2.1/anicraft-cli-0.2.1-macos-universal.zip\n- [PERSISTENCE]: The skill instructs the user or agent to modify shell profile files to maintain system PATH modifications across sessions.\n
- File: SKILL.md\n
- Evidence: export PATH="$HOME/.local/bin:$PATH" (put it in your shell profile to keep it).\n- [PRIVILEGE_ESCALATION]: The provided installation script uses the install command with specific mode bits to set executable permissions on downloaded binaries.\n
- File: SKILL.md\n
- Evidence: install -m 0755 "anicraft-cli-$APP_VER-macos-universal/anicraft-cli" "$HOME/.local/bin/anicraft-cli"\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external animation documents and vector files, which can influence the agent's behavior through the tool's 240 commands.\n
- Ingestion points: SKILL.md (references to opening .anicraft, .fla, .xfl, and SVG files).\n
- Boundary markers: Absent.\n
- Capability inventory: 240 engine commands, file writing, and external command execution (ffmpeg).\n
- Sanitization: Partial; the engine requires explicit confirmation for destructive actions.
Recommendations
- AI detected serious security threats
Audit Metadata