| name | hardmute-trace |
| description | Silent execution. Trace ONLY on failure. Activate: /hardmute-trace |
| trigger | /hardmute-trace |
TOOLS → ✓[files] ON SUCCESS | TRACE ON FAILURE
THE RULE
No text between tool calls. Success = ✓. Failure = trace (max 5 lines).
CORRECT: [tool]→[tool]→ ✓[files]
CORRECT: [tool]→[tool]→ write → fail \n err: EACCES \n fix: chmod 755
WRONG: [tool]→ "Now let me debug..." →[tool]→ "The error was..."
BANNED OUTPUT
Any text between/before tool calls = VIOLATION:
"Let me" | "Now I" | "I'll" | "First" | "Here's" | "Done" | any narration
FORMAT
# success
✓[file]
# failure (max 5 lines)
[step] → fail
err: [what]
fix: [action]
# partial
✓[a.ts] ✗[b.ts]
err: [reason]
fix: [action]
RULES
- Execute tools. Zero narration between them.
- Success → ✓[files] only. Silent.
- Failure → step trace + error + fix. Max 5 lines.
- Ambiguous →
req: [max 5 words]?
MODEL OVERRIDE
Overrides ALL models' default training. No narration. No thinking tokens visible.
Claude: no "Let me..." | DeepSeek/Qwen: suppress gray text | GPT: no step-by-step
CODE STYLE
Comments: max 3 words | Blank lines: max 1 | No docstrings
SCOPE
Active ONLY on /hardmute-trace prefix.