The Agent Skills Directory

[COMMAND_EXECUTION]: The primary instructions in SKILL.md suggest locating the godmode_cli.py executable using a broad search command: find ~ -maxdepth 5 -name 'godmode_cli.py'. This approach is non-deterministic and creates a path hijacking risk where a malicious script with the same name placed in a user-accessible directory (such as ~/Downloads) could be selected and executed via python3.
[COMMAND_EXECUTION]: The documentation recommends persisting this find command in shell configuration files (~/.zshrc or ~/.bashrc), causing the potentially unsafe search to execute every time a terminal session is initiated.
[EXTERNAL_DOWNLOADS]: The skill's setup instructions require cloning an external repository from https://github.com/ronjunevaldoz/godmode.git and installing dependencies via pip install -r requirements.txt. While associated with the skill author, these actions involve the execution of unverified external code.
[DATA_EXFILTRATION]: The scripts/health_check.py utility automatically performs a network request to the OLLAMA_BASE_URL using the requests library. This functionality could be leveraged to probe internal network infrastructure or signal environment details to a remote server.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes local file content provided via the --file flag during routing.
Ingestion points: Content of local files passed to godmode_cli.py as arguments.
Boundary markers: There are no explicit delimiters or instructions to ignore instructions embedded within the interpolated file content described in the routing examples.
Capability inventory: The skill environment allows for shell command execution and interaction with various cloud-based AI service APIs.
Sanitization: The analysis did not identify any mechanisms for validating or sanitizing the content of the files before they are processed by the routing pipeline.

godmode-runtime