dsh-perm-guard

Heavy users of DeepSeek Harness web know the pain: workspace-write pops a confirmation dialog for every cross-directory edit and Git commit, while danger-full-access hands over the keys with no guardrails at all. dsh-perm-guard fills the gap between those two extremes. It intercepts every approval request before the host prompt fires, silently auto-approves everyday operations like file edits, Git local commands, and build-test cycles, and still demands human sign-off for deletes, disk operations, privilege escalation, and curl-pipe-sh patterns. You reclaim your flow without sacrificing a safety net.

Two switchable modes cover different risk appetites. Standard auto-approves inside your trust directories (workspace, siblings, and a custom list) and prompts outside them or on risky categories. Aggressive drops the location restriction entirely and only gates destructive operations. On top of the mode defaults, eleven per-category tri-state switches (auto, ask, deny) let you fine-tune any single category to your preference. Every decision is logged with a timestamp and command summary, and the configuration persists across restarts with zero external dependencies.

It is built for power developers who live in dsh web all day and have grown tired of clicking through dozens of approval cards, yet want a hard floor under the automation. The classifier covers bash, PowerShell, and file-tool entry points; compound commands are split per sub-command and the strictest verdict wins; anything the parser cannot understand always falls back to asking. Whether you are on macOS, Linux, or Windows, it slots in as a single plugin and never changes the host behaviour once you toggle it off.