Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Prepare a journalist to interview a source or subject — including hostile or accountability interviews. Use when a reporter needs to prep an interview with a source, plan questions for a subject, handle an on-the-record accountability interview, or get a reluctant person to talk. Produces a question plan sequenced from rapport to the hard asks, ground-rules handling (on/off record, attribution), techniques for evasive or hostile subjects, and a capture plan. Distinct from expert-interview-prep (
.claude/skills/mohitagw15856-source-interview-prep/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 71% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 25% | 0% |
A journalistic interview isn't a friendly chat or a research call — it's an on-the-record exchange where the subject may have every reason not to answer straight. This skill preps the interview to get truthful, usable, attributable answers: rapport first, the hard questions sequenced so they don't end the conversation early, and a plan for the dodge.
Given the subject and the story, write the full prep — infer where the subject is cooperative vs. defensive and sequence accordingly. Always handle ground rules explicitly; never counsel deception or entrapment.
Ask for (if not provided, else infer and label):
On-the-record vs. background vs. off-the-record vs. not-for-attribution — define them, agree them up front, and note how you'll handle mid-interview attempts to go off-record.
Techniques for the evasive or hostile subject: the follow-up that closes the gap, repeating the unanswered question, using documents/quotes to press, silence, and staying neutral rather than arguing.
Recording (and consent/legal note — recording-consent laws vary by jurisdiction), notes, and how you'll verify quotes.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 36,825 | 36,513 | -1% | 1 | 1 | 0% | 4,742 | 4,892 | +3% | 0 | 0 | — |
case-02 | fail→pass | 23,447 | 26,534 | +13% | 1 | 1 | 0% | 3,659 | 5,322 | +45% | 0 | 0 | — |
case-03 | fail→pass | 23,338 | 22,856 | -2% | 1 | 1 | 0% | 2,471 | 3,882 | +57% | 0 | 0 | — |
case-04 | fail→pass | 22,604 | 26,878 | +19% | 1 | 1 | 0% | 2,554 | 4,097 | +60% | 0 | 0 | — |
case-05 | fail→fail | 23,742 | 26,611 | +12% | 1 | 1 | 0% | 2,729 | 3,912 | +43% | 0 | 0 | — |
case-06 | pass→pass | 21,849 | 18,484 | -15% | 1 | 1 | 0% | 2,148 | 3,330 | +55% | 0 | 0 | — |
case-07 | fail→pass | 38,347 | 16,282 | -58% | 1 | 1 | 0% | 1,761 | 3,010 | +71% | 0 | 0 | — |
case-08 | fail→pass | 22,282 | 23,825 | +7% | 1 | 1 | 0% | 2,751 | 3,425 | +25% | 0 | 0 | — |
case-09 | fail→pass | 20,335 | 26,048 | +28% | 1 | 1 | 0% | 2,236 | 3,890 | +74% | 0 | 0 | — |
case-10 | fail→pass | 19,579 | 23,561 | +20% | 1 | 1 | 0% | 2,306 | 3,428 | +49% | 0 | 0 | — |
case-11 | pass→pass | 19,484 | 26,520 | +36% | 1 | 1 | 0% | 2,233 | 3,860 | +73% | 0 | 0 | — |
case-12 | pass→pass | 19,858 | 24,087 | +21% | 1 | 1 | 0% | 2,223 | 3,576 | +61% | 0 | 0 | — |
case-13 | pass→pass | 21,842 | 27,160 | +24% | 1 | 1 | 0% | 2,563 | 3,886 | +52% | 0 | 0 | — |
case-14 | pass→pass | 17,213 | 29,344 | +70% | 1 | 1 | 0% | 2,521 | 4,093 | +62% | 0 | 0 | — |
case-15 | pass→pass | 20,875 | 31,913 | +53% | 1 | 1 | 0% | 2,202 | 4,143 | +88% | 0 | 0 | — |
case-16 | pass→pass | 20,489 | 16,202 | -21% | 1 | 1 | 0% | 2,270 | 3,122 | +38% | 0 | 0 | — |
case-17 | pass→pass | 17,581 | 25,103 | +43% | 1 | 1 | 0% | 2,109 | 3,745 | +78% | 0 | 0 | — |
case-18 | pass→pass | 21,363 | 12,837 | -40% | 1 | 1 | 0% | 1,946 | 2,020 | +4% | 0 | 0 | — |
case-19 | pass→pass | 13,542 | 20,599 | +52% | 1 | 1 | 0% | 2,301 | 3,534 | +54% | 0 | 0 | — |
case-20 | pass→pass | 22,371 | 17,980 | -20% | 1 | 1 | 0% | 2,789 | 3,355 | +20% | 0 | 0 | — |
case-21 | fail→fail | 3,235 | 18,194 | +462% | 1 | 1 | 0% | 352 | 3,459 | +883% | 0 | 0 | — |
case-22 | pass→fail | 21,720 | 31,896 | +47% | 1 | 1 | 0% | 3,222 | 4,760 | +48% | 0 | 0 | — |
case-23 | pass→fail | 13,748 | 23,837 | +73% | 1 | 1 | 0% | 2,293 | 4,287 | +87% | 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. 23 cases were attempted. The headline lift of +22 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are 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.