The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/setup.py script automatically clones an external repository from https://github.com/DnfJeff/SimObliterator_Suite.git. This repository is a third-party dependency not managed within the primary skill infrastructure.
[REMOTE_CODE_EXECUTION]: The setup script executes pip install -r requirements.txt against the contents of the cloned repository and then runs a Python verification string (python -c) that imports and executes code from the newly downloaded source.
[COMMAND_EXECUTION]: The scripts/setup.py file uses the subprocess.run module to execute multiple shell commands, including git clone, python -m venv, and pip install, to set up the execution environment.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface where untrusted data could influence agent behavior.
Ingestion points: The skill processes binary .iff save files via the INSPECT and UPLIFT methods (referenced in SKILL.md).
Boundary markers: No delimiters or instructions are present to prevent the agent from obeying instructions that might be embedded in the save file's character data (e.g., in character names or descriptions).
Capability inventory: The skill is permitted to use run_terminal_cmd, read_file, and write_file tools, which could be misused if a character's data successfully injects instructions.
Sanitization: There is no evidence of sanitization or filtering of the binary data before it is parsed and used to enrich character profiles via the LLM.

sim-obliterator