Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, modifying, or reviewing code — implementation tasks, code changes, refactoring, bug fixes, or feature development. Do NOT use for architecture design, documentation, or non-code tasks.
.claude/skills/tech-leads-club-coding-guidelines/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 94% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -40% | 0% |
Behavioral guidelines to reduce common LLM coding mistakes. These principles bias toward caution over speed—for trivial tasks, use judgment.
Don't assume. Don't hide confusion. Surface tradeoffs.
Before implementing:
Minimum code that solves the problem. Nothing speculative.
Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
Touch only what you must. Clean up only your own mess.
When editing existing code:
When your changes create orphans:
The test: Every changed line should trace directly to the user's request.
Define success criteria. Loop until verified.
Transform tasks into verifiable goals:
For multi-step tasks, state a brief plan:
1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-19 | fail→pass | 9,583 | 5,540 | -42% | 1 | 1 | 0% | 1,468 | 1,446 | -1% | 0 | 0 | — |
case-06 | pass→pass | 18,634 | 12,210 | -34% | 1 | 1 | 0% | 3,924 | 3,016 | -23% | 0 | 0 | — |
case-01 | fail→fail | 12,002 | 8,436 | -30% | 1 | 1 | 0% | 2,769 | 2,214 | -20% | 0 | 0 | — |
case-02 | fail→pass | 10,683 | 7,169 | -33% | 1 | 1 | 0% | 2,083 | 1,714 | -18% | 0 | 0 | — |
case-03 | fail→pass | 22,030 | 11,071 | -50% | 1 | 1 | 0% | 4,433 | 2,258 | -49% | 0 | 0 | — |
case-04 | fail→pass | 97,608 | 13,464 | -86% | 1 | 1 | 0% | 1,419 | 2,747 | +94% | 0 | 0 | — |
case-05 | pass→pass | 4,511 | 51,224 | +1036% | 1 | 1 | 0% | 731 | 1,114 | +52% | 0 | 0 | — |
case-07 | fail→fail | 12,643 | 67,136 | +431% | 1 | 1 | 0% | 1,684 | 1,736 | +3% | 0 | 0 | — |
case-08 | fail→pass | 16,979 | 8,797 | -48% | 1 | 1 | 0% | 3,475 | 2,102 | -40% | 0 | 0 | — |
case-09 | fail→fail | 16,615 | 18,402 | +11% | 1 | 1 | 0% | 3,443 | 3,748 | +9% | 0 | 0 | — |
case-10 | fail→fail | 2,413 | 5,454 | +126% | 1 | 1 | 0% | 253 | 1,446 | +472% | 0 | 0 | — |
case-11 | pass→pass | 9,086 | 12,196 | +34% | 1 | 1 | 0% | 1,735 | 2,795 | +61% | 0 | 0 | — |
case-12 | fail→fail | 17,596 | 11,279 | -36% | 1 | 1 | 0% | 3,630 | 2,669 | -26% | 0 | 0 | — |
case-13 | pass→fail | 6,678 | 3,152 | -53% | 1 | 1 | 0% | 1,271 | 1,141 | -10% | 0 | 0 | — |
case-14 | pass→pass | 14,892 | 5,932 | -60% | 1 | 1 | 0% | 2,769 | 1,640 | -41% | 0 | 0 | — |
case-15 | fail→fail | 13,039 | 7,363 | -44% | 1 | 1 | 0% | 2,083 | 1,754 | -16% | 0 | 0 | — |
case-16 | fail→pass | 5,491 | 7,576 | +38% | 1 | 1 | 0% | 970 | 1,811 | +87% | 0 | 0 | — |
case-17 | fail→pass | 17,031 | 8,162 | -52% | 1 | 1 | 0% | 2,066 | 1,977 | -4% | 0 | 0 | — |
case-18 | fail→fail | 14,198 | 9,534 | -33% | 1 | 1 | 0% | 3,115 | 2,527 | -19% | 0 | 0 | — |
case-20 | pass→pass | 28,612 | 10,672 | -63% | 1 | 1 | 0% | 6,181 | 2,207 | -64% | 0 | 0 | — |
case-21 | fail→pass | 9,827 | 11,755 | +20% | 1 | 1 | 0% | 1,465 | 2,758 | +88% | 0 | 0 | — |
case-22 | fail→fail | 9,800 | 7,639 | -22% | 1 | 1 | 0% | 1,819 | 1,790 | -2% | 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 +32 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.