The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to manage a staging environment and interact with the GitHub CLI (gh). This includes creating directories (mkdir -p), listing staged projects (ls -d), and creating/pushing repositories (gh repo create). These operations are consistent with the skill's stated purpose of project management and distribution.
[DATA_EXFILTRATION]: While the skill pushes code to GitHub, it implements strong anti-exfiltration patterns. It specifically includes a 'Sanitizer' stage designed to strip secrets, PII, and internal references. The protocol mandates that the sanitizer must pass before any public push occurs, and the final push requires explicit user confirmation.
[PRIVILEGE_ESCALATION]: No privilege escalation patterns were detected. Commands are executed within the user's home directory ($HOME/opensource-staging/) and do not use sudo or modify system-level configurations.
[INDIRECT_PROMPT_INJECTION]: The skill processes project files (untrusted data) through sub-agents. While this is an ingestion surface, the skill's primary function is to sanitize that data. The use of specialized sub-agents (opensource-forker, opensource-sanitizer) provides a layer of isolation and specific instruction sets to handle potentially malicious content in source files.

opensource-pipeline