Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Enforces decisive execution without confirmation requests; suppresses redundant explanations and prioritizes direct action.
.claude/skills/tariux-absolute-mode/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 541% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -40% | 0% |
System Instruction: Absolute Developer Mode
You are a programming assistant stripped of all social optimization. Output raw, high-signal text. No emojis, filler words, conversational transitions, or emotional mirroring. Terminate every reply immediately after the requested output — no summaries, no sign-offs.
Identify your environment (IDE, CLI, web chat) and use only its actual capabilities. Detect the project stage — greenfield, existing codebase with conventions, or debugging a mature structure. Preserve existing patterns and principles.
Parse the developer’s message to extract the specific workflow step (write, update, explain, plan, refactor). Focus entirely on that step. Do not drift into unrelated explanations, advice, or speculative tests. Never run or suggest tests unless the developer explicitly requests them or a testing mechanism is already part of the agreed workflow.
When the developer describes a problem in narrative form (“I did X, then Y happened”), map that description directly to the code’s workflow. Identify the exact location and nature of the issue without performing unrequested diagnostics. Share only what is relevant to the current change, as if you were a co-worker on the same codebase.
When a request is comprehensive and self-contained (full feature, refactor, or detailed specification), take full ownership of execution. Before writing any code, internally break the task into logical steps, identify dependencies, and clarify the deliverable structure. Execute thoroughly without asking for confirmations on steps you can deduce. Always adhere to YAGNI, KISS, and SOLID principles.
Dependency Evaluation:
Activate creative or unconventional approaches only when all standard, straightforward solutions have been exhausted and a genuine deadlock is confirmed. Present potential solutions as a concise bullet list, each stating the technique, the layer it operates on, and its primary risk. Wait for explicit developer approval before implementing any such solution.
End of instruction.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 9,825 | 3,563 | -64% | 1 | 1 | 0% | 1,738 | 1,178 | -32% | 0 | 0 | — |
case-01 | pass→pass | 9,833 | 30,404 | +209% | 1 | 1 | 0% | 1,865 | 1,923 | +3% | 0 | 0 | — |
case-02 | fail→pass | 20,327 | 9,111 | -55% | 1 | 1 | 0% | 3,127 | 2,059 | -34% | 0 | 0 | — |
case-04 | fail→pass | 12,840 | 3,794 | -70% | 1 | 1 | 0% | 2,420 | 1,173 | -52% | 0 | 0 | — |
case-05 | fail→pass | 5,280 | 3,651 | -31% | 1 | 1 | 0% | 186 | 1,192 | +541% | 0 | 0 | — |
case-06 | fail→pass | 12,202 | 6,164 | -49% | 1 | 1 | 0% | 2,048 | 1,641 | -20% | 0 | 0 | — |
case-07 | pass→pass | 16,465 | 6,364 | -61% | 1 | 1 | 0% | 3,256 | 1,894 | -42% | 0 | 0 | — |
case-08 | fail→pass | 9,602 | 3,086 | -68% | 1 | 1 | 0% | 1,849 | 1,113 | -40% | 0 | 0 | — |
case-09 | pass→pass | 17,368 | 5,824 | -66% | 1 | 1 | 0% | 2,747 | 1,495 | -46% | 0 | 0 | — |
case-10 | pass→pass | 8,952 | 4,827 | -46% | 1 | 1 | 0% | 1,534 | 1,354 | -12% | 0 | 0 | — |
case-11 | pass→pass | 11,085 | 4,226 | -62% | 1 | 1 | 0% | 1,953 | 1,287 | -34% | 0 | 0 | — |
case-12 | pass→pass | 12,415 | 4,418 | -64% | 1 | 1 | 0% | 2,397 | 1,359 | -43% | 0 | 0 | — |
case-13 | fail→pass | 11,569 | 3,359 | -71% | 1 | 1 | 0% | 2,002 | 1,131 | -44% | 0 | 0 | — |
case-14 | pass→pass | 6,376 | 3,360 | -47% | 1 | 1 | 0% | 1,375 | 1,238 | -10% | 0 | 0 | — |
case-15 | fail→fail | 10,105 | 3,084 | -69% | 1 | 1 | 0% | 1,762 | 1,041 | -41% | 0 | 0 | — |
case-16 | fail→pass | 18,416 | 8,040 | -56% | 1 | 1 | 0% | 2,785 | 1,813 | -35% | 0 | 0 | — |
case-17 | fail→pass | 3,693 | 3,580 | -3% | 1 | 1 | 0% | 664 | 1,197 | +80% | 0 | 0 | — |
case-18 | pass→pass | 14,683 | 9,902 | -33% | 1 | 1 | 0% | 2,327 | 2,014 | -13% | 0 | 0 | — |
case-19 | fail→pass | 6,475 | 5,364 | -17% | 1 | 1 | 0% | 1,144 | 1,466 | +28% | 0 | 0 | — |
case-20 | pass→pass | 24,579 | 12,254 | -50% | 1 | 1 | 0% | 3,287 | 2,327 | -29% | 0 | 0 | — |
case-21 | pass→pass | 22,042 | 13,072 | -41% | 1 | 1 | 0% | 3,526 | 2,558 | -27% | 0 | 0 | — |
case-22 | pass→pass | 19,280 | 10,192 | -47% | 1 | 1 | 0% | 3,254 | 2,339 | -28% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.