Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Prepare for and script a hard conversation — conflict, bad news, a boundary, an apology. Use when asked to prepare for a difficult conversation, address a conflict, deliver bad news, confront a colleague, or have a hard talk with a manager/report/peer. Produces a prep brief — the real goal, the other side's likely view, an opening line, the key points, anticipated reactions with responses, and the outcome you want.
.claude/skills/mohitagw15856-difficult-conversation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 18% | 0% |
The conversations we avoid are usually the ones that matter most — and we botch them by winging it or over-rehearsing into a script that shatters on first contact. This skill preps the hard talk the way the research says works: get clear on the actual goal, understand the other person's story, open without triggering defensiveness, and plan for their reactions — so you go in calm and come out with the relationship intact.
Ask for these only if they aren't already provided:
1. Your real goal — name it plainly (and the un-goal — e.g. "not to win, but to change X"). Conversations go wrong when the unspoken goal is to be proven right.
2. Their story — how they likely see it and what they need to feel (heard, respected, safe). You can't move someone you haven't understood.
3. Open — a specific opening line that states the issue from the facts + your impact, not blame ("When the deadline slipped, I was left explaining it to the client" — not "You always miss deadlines"). The first 30 seconds set the tone.
4. Key points — the 2–3 things you must convey, each separating observation from story/judgement.
5. Likely reactions → your response — defensiveness, deflection, emotion, counterattack — and a calm, non-escalating reply prepared for each.
| If they… | You respond… | |---|---|
6. Land it — the ask or agreement you want, and how to close on a concrete next step.
Stance note — stay curious, not certain; aim for a shared understanding, not a verdict.
Crucial Conversations (Patterson et al.) and Difficult Conversations (Stone, Patton, Heen) — facts vs. story, the third story, safety.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | pass→pass | 14,767 | 12,905 | -13% | 1 | 1 | 0% | 2,439 | 3,010 | +23% | 0 | 0 | — |
case-01 | fail→pass | 16,760 | 17,143 | +2% | 1 | 1 | 0% | 2,597 | 3,422 | +32% | 0 | 0 | — |
case-02 | fail→pass | 14,936 | 11,064 | -26% | 1 | 1 | 0% | 2,397 | 2,764 | +15% | 0 | 0 | — |
case-03 | pass→pass | 10,526 | 10,362 | -2% | 1 | 1 | 0% | 1,556 | 2,319 | +49% | 0 | 0 | — |
case-04 | pass→pass | 11,076 | 8,999 | -19% | 1 | 1 | 0% | 1,835 | 2,387 | +30% | 0 | 0 | — |
case-05 | pass→pass | 8,116 | 7,390 | -9% | 1 | 1 | 0% | 1,354 | 1,781 | +32% | 0 | 0 | — |
case-06 | fail→pass | 14,225 | 15,975 | +12% | 1 | 1 | 0% | 2,223 | 3,262 | +47% | 0 | 0 | — |
case-07 | pass→pass | 13,900 | 17,644 | +27% | 1 | 1 | 0% | 1,951 | 3,300 | +69% | 0 | 0 | — |
case-08 | pass→pass | 13,883 | 12,016 | -13% | 1 | 1 | 0% | 2,327 | 2,620 | +13% | 0 | 0 | — |
case-09 | pass→pass | 15,294 | 12,727 | -17% | 1 | 1 | 0% | 2,532 | 2,998 | +18% | 0 | 0 | — |
case-10 | pass→pass | 14,501 | 13,586 | -6% | 1 | 1 | 0% | 2,404 | 3,144 | +31% | 0 | 0 | — |
case-12 | fail→pass | 15,995 | 14,035 | -12% | 1 | 1 | 0% | 2,462 | 3,182 | +29% | 0 | 0 | — |
case-13 | fail→pass | 17,040 | 14,854 | -13% | 1 | 1 | 0% | 2,521 | 2,969 | +18% | 0 | 0 | — |
case-14 | fail→pass | 10,401 | 2,725 | -74% | 1 | 1 | 0% | 1,769 | 1,229 | -31% | 0 | 0 | — |
case-15 | pass→pass | 16,461 | 17,174 | +4% | 1 | 1 | 0% | 2,438 | 3,160 | +30% | 0 | 0 | — |
case-16 | fail→pass | 12,497 | 13,437 | +8% | 1 | 1 | 0% | 2,130 | 2,801 | +32% | 0 | 0 | — |
case-17 | fail→pass | 17,817 | 15,758 | -12% | 1 | 1 | 0% | 2,572 | 3,311 | +29% | 0 | 0 | — |
case-18 | pass→pass | 18,516 | 20,005 | +8% | 1 | 1 | 0% | 2,662 | 3,751 | +41% | 0 | 0 | — |
case-19 | fail→pass | 15,124 | 11,443 | -24% | 1 | 1 | 0% | 2,169 | 2,519 | +16% | 0 | 0 | — |
case-20 | fail→fail | 15,254 | 15,797 | +4% | 1 | 1 | 0% | 2,267 | 3,285 | +45% | 0 | 0 | — |
case-21 | fail→pass | 12,103 | 14,215 | +17% | 1 | 1 | 0% | 1,951 | 3,042 | +56% | 0 | 0 | — |
case-22 | fail→pass | 17,918 | 13,592 | -24% | 1 | 1 | 0% | 2,503 | 2,769 | +11% | 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. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 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.