The Agent Skills Directory

[SAFE]: The skill implements a robust engineering workflow using isolated Git worktrees and parallel role-specific threads (orchestrator, implementer, tester, reviewer). This architecture ensures a clear separation of concerns and facilitates rigorous independent verification of all code changes.
[SAFE]: Included shell scripts for workspace initialization and state management (oracle_init.sh, oracle_lane_state.sh, oracle_start_run.sh) are designed with significant defensive measures. Specifically, they strictly validate file paths and prohibit operations on symlinks to mitigate risks of directory traversal and symlink-based attacks.
[SAFE]: The sandbox creation process (oracle_sandbox_worktree.sh) handles sensitive environment configuration securely. It verifies that .env files are properly Git-ignored and not tracked by version control before copying them to isolated worktrees with restricted filesystem permissions (mode 0600).
[SAFE]: To mitigate risks associated with indirect prompt injection from project data, the skill uses "narrow context packets" for sub-agents. This ensures that each thread (e.g., a tester or reviewer) receives only the specific context required for its task, effectively isolating the agent from potentially malicious instructions embedded in the codebase.

oracle