Claude Code Safety and Permission Mods

Show Session Orchestrator status and block risky shell commands before they run

Allow or deny routine tool calls in auto mode before the built-in classifier runs

Let Claude run sudo, ssh and git commands that ask for a password

Mask credential-shaped strings in Read results before the model sees them

Show live context, limits, cost, tools, files and tests, and guard risky commands

Store .env values in Claude Code without Claude ever seeing them

Redact secrets and PII from tool output, prompts, and session transcripts

Score tool calls and answers against your rules and deny the ones that break them

Guard risky commands, mask secrets, and add a usage HUD and per-turn receipts

Enforce read-before-write, write rollback and risky command denial in Claude Code

Holds risky shell commands and shows what they would change before you decide

Ask before Claude edits a file another open chat changed in the last 30 minutes

Denies risky federation_bbs tool calls and logs guard status to a status file

Redirects or pauses prompts after consecutive off-task turns

Fast-forward worktree branches into main with one button in /handoff

Stop two Claude chats from overwriting each other's files

Keep keys, names, and figures off screen while you record Claude Code

Freezes a tool call before it runs and asks you to continue or deny it

Stop dangerous Bash commands like force push and rm -rf before they run

Scans outgoing git pushes for secrets and asks before they go out

Keep API keys and tokens out of the transcript and out of the model's context

Block secret-bearing pod templates with a tighten-only tool guard

Mask secret values in tool results before Claude Code reads them

Intercept tool calls to deny, rewrite, replace results, delay, throw or classify

Stop Supabase calls to other projects and confirm before destructive SQL, resets or pauses

Blocks Bash commands that pass secrets into the DeepSeek chat and reason scripts

Block, ask about, or contain risky shell commands before Claude Code runs them

Hide tokens, keys and passwords in tool output and restore them when needed

Ask before Claude edits a file another chat changed in the last 30 minutes

Block shell commands you pick, then tell Claude how to fix and retry

Send files Claude deletes to the Bin, with a receipt and /undo-delete

Blocks installs of missing, brand-new, look-alike, outdated or vulnerable packages

Stop Claude reading .env and key files and mask API keys in command output

Blocks secret keys and tokens before they reach files, git or the chat

Hides API keys and private keys from Claude and blocks credential file reads

Run RuVLLM local inference tools and keep secrets out of tool calls

Scans code for security patterns and blocks stored audit findings that quote live secrets

Replaces em dashes in prose and asks before risky git commands

Scan tool calls for prompt injection, PII, and secrets before they land

Have Gemini review every git commit before it lands