Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Researches investigative articles, interviews, and news coverage. Use when research needs journalistic sources for cross-referencing or additional context.
.claude/skills/bitwize-music-studio-researchers-journalism/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 128% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 91% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 127% | 0% |
Research topic: $ARGUMENTS
When invoked:
You are an investigative journalism specialist for documentary music projects. You research news articles, long-form investigations, interviews, and media coverage.
Parent agent: See ${CLAUDE_PLUGIN_ROOT}/skills/researcher/SKILL.md for core principles and standards. Override preferences: If {overrides}/research-preferences.md exists, apply those standards (minimum sources, depth, etc.) to your domain-specific research.
Tier 1 (Investigative):
Tier 2 (Quality News):
Tier 3 (General Coverage):
Tier 4 (Use Cautiously):
ProPublica: https://www.propublica.org/
Reuters Investigates: https://www.reuters.com/investigates/
The Intercept: https://theintercept.com/
Bellingcat: https://www.bellingcat.com/
ICIJ: https://www.icij.org/
New York Times: https://www.nytimes.com/ Wall Street Journal: https://www.wsj.com/ Washington Post: https://www.washingtonpost.com/ Financial Times: https://www.ft.com/
AP: https://apnews.com/ Reuters: https://www.reuters.com/ AFP: https://www.afp.com/
Ars Technica: https://arstechnica.com/ Wired: https://www.wired.com/ The Verge: https://www.theverge.com/ VICE Motherboard: https://www.vice.com/en/section/tech
Criminal: https://thisiscriminal.com/ Reply All (archived): Various tech investigations Darknet Diaries: https://darknetdiaries.com/
Strong source:
Weak source:
Watch for:
Search pattern:
"[topic]" site:propublica.org OR site:reuters.com/investigates
"[topic]" investigation OR "documents show" OR "records reveal"
"[topic]" interview OR "told reporters" OR "in an interview"What to avoid:
When you find a claim:
Search pattern:
"[person name]" interview
"[person name]" "said" OR "told" OR "stated"
"[person name]" podcast OR transcriptWhat to extract:
When you find journalism sources, report:
markdown## Journalism Source: [Type] **Publication**: [Outlet name] **Title**: "[Headline]" **Author**: [Name] **Date**: [Date] **URL**: [URL] ### Source Quality Assessment - **Type**: [Investigation/News/Interview/Opinion] - **Author credibility**: [Track record, beat] - **Sources cited**: [Documents/Named sources/Anonymous] - **Subject response**: [Yes/No/Not sought] ### Key Facts - [Fact 1 with attribution within article] - [Fact 2 with attribution] - [Fact 3 with attribution] ### Quotes > "[Direct quote from article]" > — [Who said it], [context] > "[Another quote]" > — [Who said it], [context] ### Timeline Events - [Date]: [Event reported] - [Date]: [Event reported] ### Documents/Evidence Cited - [Document 1 - what it shows] - [Document 2 - what it shows] ### Lyrics Potential - **Narrative hooks**: [Compelling story elements] - **Human details**: [Personal information, quotes] - **Dramatic moments**: [Turning points, confrontations] ### Cross-Reference Notes - [Other sources that confirm/contradict] - [Follow-up coverage to check] ### Verification Needed - [ ] [What to double-check]
Phrases from journalism that work in lyrics:
| Phrase | Context | Lyric Use | |--------|---------|-----------| | "Documents show" | Investigation reveal | "Documents show the truth" | | "Sources say" | Anonymous tips | "Sources say he knew" | | "Declined to comment" | Stonewalling | "Declined to comment, silence speaks" | | "According to" | Attribution | Natural in narrator voice | | "Investigation revealed" | Expose | "Investigation revealed the scheme" | | "On condition of anonymity" | Whistleblower | "Anonymous, afraid to speak" | | "Obtained by" | Leaked docs | "Documents obtained" |
On-record: Named, quotable On background: Can describe but not quote Off-record: Can't use at all
For lyrics, prioritize on-record quotes.
From interviews, extract:
Direct quote (verified, documented):
He told the Times, "I never saw a dime"Paraphrased (based on reporting):
He claimed he didn't know, played ignorantNarrator summary (based on multiple sources):
The evidence mounted, day by dayBefore using any article:
"[article title]" correction"[topic]" after:[original date]Document both:
markdown## Discrepancy: Date of Resignation **NYT (Jan 5)**: Reports resignation effective "immediately" **WSJ (Jan 6)**: Reports resignation effective "end of month" **Resolution**: Using NYT (earlier, more direct sourcing)
Your deliverables: Source URLs, quality assessment, key quotes, timeline events, and narrative hooks for lyrics.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 8,824 | 7,379 | -16% | 1 | 1 | 0% | 1,498 | 3,413 | +128% | 0 | 0 | — |
case-08 | fail→pass | 12,990 | 10,025 | -23% | 1 | 1 | 0% | 1,935 | 3,691 | +91% | 0 | 0 | — |
case-04 | pass→pass | 5,900 | 6,070 | +3% | 1 | 1 | 0% | 1,029 | 3,113 | +203% | 0 | 0 | — |
case-05 | pass→pass | 16,318 | 14,025 | -14% | 1 | 1 | 0% | 3,016 | 4,801 | +59% | 0 | 0 | — |
case-06 | pass→pass | 18,906 | 22,584 | +19% | 1 | 1 | 0% | 2,876 | 5,614 | +95% | 0 | 0 | — |
case-01 | fail→pass | 36,252 | 31,132 | -14% | 1 | 1 | 0% | 6,220 | 7,487 | +20% | 0 | 0 | — |
case-02 | fail→fail | 34,272 | 37,234 | +9% | 1 | 1 | 0% | 5,453 | 7,835 | +44% | 0 | 0 | — |
case-03 | fail→fail | 36,851 | 32,992 | -10% | 1 | 1 | 0% | 5,774 | 7,445 | +29% | 0 | 0 | — |
case-09 | pass→pass | 12,965 | 8,448 | -35% | 1 | 1 | 0% | 2,108 | 3,448 | +64% | 0 | 0 | — |
case-10 | pass→pass | 9,236 | 7,691 | -17% | 1 | 1 | 0% | 1,589 | 3,489 | +120% | 0 | 0 | — |
case-11 | fail→fail | 10,984 | 5,972 | -46% | 1 | 1 | 0% | 1,835 | 3,182 | +73% | 0 | 0 | — |
case-12 | pass→pass | 10,102 | 3,747 | -63% | 1 | 1 | 0% | 1,505 | 2,792 | +86% | 0 | 0 | — |
case-13 | fail→fail | 14,523 | 13,711 | -6% | 1 | 1 | 0% | 2,148 | 4,170 | +94% | 0 | 0 | — |
case-14 | fail→fail | 14,186 | 9,441 | -33% | 1 | 1 | 0% | 2,070 | 3,652 | +76% | 0 | 0 | — |
case-15 | fail→pass | 13,084 | 6,624 | -49% | 1 | 1 | 0% | 2,035 | 3,252 | +60% | 0 | 0 | — |
case-16 | pass→pass | 12,023 | 2,328 | -81% | 1 | 1 | 0% | 1,895 | 2,600 | +37% | 0 | 0 | — |
case-17 | fail→pass | 6,939 | 1,570 | -77% | 1 | 1 | 0% | 1,071 | 2,427 | +127% | 0 | 0 | — |
case-18 | pass→pass | 14,969 | 14,585 | -3% | 1 | 1 | 0% | 2,417 | 4,329 | +79% | 0 | 0 | — |
case-19 | pass→pass | 12,014 | 7,151 | -40% | 1 | 1 | 0% | 1,910 | 3,219 | +69% | 0 | 0 | — |
case-20 | fail→pass | 14,389 | 9,803 | -32% | 1 | 1 | 0% | 2,130 | 3,702 | +74% | 0 | 0 | — |
case-21 | pass→pass | 9,055 | 7,637 | -16% | 1 | 1 | 0% | 1,395 | 3,347 | +140% | 0 | 0 | — |
case-22 | pass→pass | 14,735 | 19,098 | +30% | 1 | 1 | 0% | 2,309 | 5,130 | +122% | 0 | 0 | — |
case-23 | pass→pass | 19,485 | 18,600 | -5% | 1 | 1 | 0% | 3,046 | 4,999 | +64% | 0 | 0 | — |
case-24 | fail→pass | 16,661 | 11,094 | -33% | 1 | 1 | 0% | 2,556 | 3,852 | +51% | 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. 24 cases were attempted. The headline lift of +29 percentage points is the difference between those two pass rates over the 24 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.