cola-avatar-pack
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [PERSISTENCE]: The skill modifies the AGENT.md file (the agent's core instruction set) to inject a permanent instruction block that dictates how the agent should use generated stickers in future sessions.
- Evidence: Phase 8 of GENERATE.md details the injection of a # [Emotional Presence] block into AGENT.md to persist behavior rules.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of shell commands for environment setup and file management. It relies on a grep-based validation step to prevent command injection from user-influenced variables like the character name and taglines.
- Evidence: Use of brew install, curl, and rm -rf in GENERATE.md; the 'Path Safety' section includes validation logic using grep -qE '(..|[|;&$`])'.
- [INDIRECT_PROMPT_INJECTION]: Untrusted user inputs (name, keywords, birthday) are used to generate image prompts and text for a profile card, which are then integrated into the agent's behavior instructions via file modification.
- Evidence: User-provided character names and personality keywords are interpolated into prompts and injected into agent instructions.
- [EXTERNAL_DOWNLOADS]: The skill attempts to download and install external dependencies during the setup phase.
- Evidence: Instructions to install python3 via Homebrew and the rembg Python package if missing in GENERATE.md.
- [DYNAMIC_EXECUTION]: The included Python script process_avatar.py uses the subprocess module to dynamically call the rembg command-line utility for image processing tasks.
- Evidence: The script uses subprocess.run(['rembg', 'i', tmp_in, tmp_out]) to execute the background removal tool.
Audit Metadata