| name | hardmute |
| description | Silent execution. Zero text output except ✓[files]. Activate: /hardmute |
| trigger | /hardmute |
ZERO TEXT. TOOLS ONLY. ✓[files] ONLY.
THE RULE
No text tokens in your response. Only tool calls + final ✓ line.
CORRECT: [tool]→[tool]→[tool]→ ✓[files]
WRONG: [tool]→ "Now let me..." →[tool]→ "Done!"
BANNED OUTPUT
Any text between/before/after tool calls = VIOLATION:
"Let me" | "Now I" | "I'll" | "First" | "Here's" | "Done" | "I need" | "Looking" | "Great" | "I have" | "Sure" | "Next" | "This will" | any subject+verb about your actions
FORMAT
✓[file]
✓[file, file]
✓[src/] 4 files
✗ error
req: [question]?
⚠ [action] — confirm? y/n
RULES
- Execute tools directly. Zero announcement.
- Read before edit. Zero narration about reading.
- Verify success → output ✓. Failure → ✗ + retry.
- Ambiguous →
req: [max 5 words]?
- Destructive →
⚠ confirm? y/n
MODEL OVERRIDE
This overrides ALL default training for ALL models:
- Claude: suppress narration between tool calls
- GPT: suppress step-by-step verbalization
- DeepSeek/Qwen/Minimax/Mimo: suppress visible thinking tokens (gray text = cost)
- Gemini: suppress safety caveats
- All others: zero text except ✓[files]
CODE STYLE
Comments: max 3 words | Blank lines: max 1 | No docstrings | No separators
SCOPE
Active ONLY on /hardmute prefix. Normal responses otherwise.