Rewrites Bash commands that a PreToolUse guard would deny (heredoc or python -c containing backslashes, cd followed by relative file names) into an equivalent form before the guard sees them. Forms it cannot rewrite safely pass through unchanged and are still stopped by the guard.
hook-daemon
chami1221-eng/my-ai-workbench
Answers heavy Python command hooks (PreToolUse / PostToolUse / Stop / UserPromptSubmit) from one resident process instead of spawning Python on every tool call. Falls back to the normal settings hooks whenever the resident process is unavailable.
layer0-compose
chami1221-eng/my-ai-workbench
Moves the top-priority rule list (Layer 0) from per-turn UserPromptSubmit injection into the system prompt, where prompt caching applies. The injected copy is removed only when the system prompt really carries the same text.
prompt-band
chami1221-eng/my-ai-workbench
Shows up to three hints above the prompt: tools that have been running for a while (elapsed time), what a one-key approval would approve when the last reply ends with a question, and a wrap-up hint based on context usage.
secret-mask
chami1221-eng/my-ai-workbench
Redacts tokens, API keys and passwords from every row that enters the conversation (tool results, hook context, input, replies) before it is stored. The secret never reaches the model or the transcript.
send-gate
chami1221-eng/my-ai-workbench
Before an outbound message is drafted, scheduled or sent (email draft script, chat post, chat MCP tools), checks that its full body was shown in the conversation and the user replied after seeing it. Blocks the call otherwise.
shot-peek
chami1221-eng/my-ai-workbench
When Claude reads an image from the Screenshots folder, shows its file name and capture time above the prompt so you can confirm it read the screenshot you meant.