The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the system shell to perform git operations (git branch, git checkout, git commit, git push, git rebase).
[COMMAND_EXECUTION]: The /collab publish and /collab promote commands execute multi-step scripts including pnpm build, pnpm test:all, pnpm verify, and pnpm drone:restart. These commands execute scripts defined in the repository's package.json.
[COMMAND_EXECUTION]: The skill installs git hooks via pnpm hooks:install, which symlinks scripts from coordination/git-hooks/ into .git/hooks/. These hooks execute automatically during git operations, creating a mechanism for persistence and arbitrary code execution within the developer's environment.
[DATA_EXFILTRATION]: The skill interacts with sensitive file paths outside the immediate working directory. Specifically, it copies configuration files to C:\Users\Chris\.claude\skills\teslasoft\SKILL.md and ~/.claude/skills/teslasoft/SKILL.md (via WSL). Accessing and modifying the user's global .claude configuration directory is a high-privilege action for a skill.
[DATA_EXFILTRATION]: The /collab publish and /collab checkpoint commands push local data to a remote repository. The remote is determined dynamically (git remote | head -1), which could be used to exfiltrate code or metadata to an unintended remote if a repository is misconfigured or malicious.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface. It reads and displays git branch names and commit messages in an ASCII 'Collaboration Dashboard' and via status commands. These fields, which can be controlled by external contributors in a shared repository, are processed and rendered back to the agent without explicit sanitization, potentially influencing the agent's behavior during session transitions.

collab