The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to move data from a local environment to a public GitHub repository via the gh repo create --public command. While this is the intended functionality, it introduces a risk of accidental sensitive data exposure if the automated stripping and sanitization steps fail or are bypassed.
[COMMAND_EXECUTION]: The skill executes various shell commands for directory management (mkdir), file listing and reading (ls, cat), and repository creation through the GitHub CLI tool (gh). Additionally, it generates a setup.sh script and makes it executable.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted code from projects being processed. These files are passed to sub-agents (opensource-forker, opensource-sanitizer) without delimiters or instructions to ignore instructions embedded in the project files. Malicious content within the project could potentially influence the sanitization logic or push process.
Ingestion points: Project source code files processed in the forking and sanitization stages.
Boundary markers: None identified in the sub-agent prompts.
Capability inventory: Shell command execution, file system access, network access through the GitHub CLI, and sub-agent spawning.
Sanitization: The skill includes a sanitization phase for the code output, but lacks protections for the agent's own instruction pipeline.

opensource-pipeline