| name | nci-manipulation-analysis |
| description | Analyzes content for manipulation techniques using the NCI (Narrative Credibility Index) Protocol. Detects emotional manipulation, suspicious timing, uniform messaging, tribal division, and missing information across 20 categories. Use when asked to analyze content for manipulation, propaganda, disinformation patterns, or when user provides a URL or text asking "is this manipulative?", "analyze this for bias", "check for propaganda", or similar requests. |
NCI Manipulation Analysis
Pattern-based manipulation detection that identifies how content tries to influence you, not whether claims are factually true. Manipulation techniques leave fingerprints regardless of underlying accuracy.
Quick Start
For Text Content
- Read the content provided by user
- Apply 20-category analysis (see references/categories.md)
- Calculate composite factors and overall score (see references/scoring.md)
- Check deep research triggers - if score > 40 or key categories elevated, verify claims
- Generate dual perspectives
- Output report in requested format
For URLs
- Use
WebFetchto retrieve content from URL - Extract main article/post text
- Proceed with text analysis workflow
- Note source metadata (publication, date, author)
- If triggers met: Use
fact-checkeragent to verify key claims
First Principles (Summary)
The NCI Protocol is grounded in these principles (see agents/perspective-generator.md for full version):
- Evidence over authority - Evaluate patterns in content, not source reputation
- Steel-man interpretation - Present strongest version of each perspective
- Atomic decomposition - Break claims into smallest verifiable units
- Source agnosticism - Apply identical standards regardless of source alignment
- Bidirectional beneficiary analysis - Ask who benefits if believed AND if dismissed
- Pattern vs. Intent - Focus on techniques; deep research evidence can inform motives
These principles ensure fair, consistent analysis across all content regardless of political or ideological alignment.
Workflow
Progress:
- [ ] 1. Input Processing (text or URL)
- [ ] 2. Score all 20 categories (1-5 scale each)
- [ ] 3. Calculate 5 composite factors
- [ ] 4. Calculate overall score (0-100)
- [ ] 5. Check deep research triggers (see Step 5)
- [ ] 6. Generate perspectives (manipulative + legitimate)
- [ ] 7. Output report
Step 1: Input Processing
For direct text:
INPUT TYPE: Text
LENGTH: [word count]
CONTEXT PROVIDED: [any user context]
For URLs:
INPUT TYPE: URL
URL: [url]
Fetching content with WebFetch...
EXTRACTED: [article title, publication, date if available]
When processing URLs, also check:
- Publication reputation
- Author credentials (if available)
- Publication date and timeliness
Step 2: Score All 20 Categories
For each category, provide:
CATEGORY #[N]: [Name]
Score: [1-5]
Evidence: [Specific quotes/patterns from content]
Confidence: [LOW/MED/HIGH]
See references/categories.md for detailed category definitions and scoring criteria.
Detection signals to look for:
| Signal Type | Examples |
|---|---|
| Emotional vocabulary | fear, outrage, danger, threat, shocking |
| Urgency language | immediately, urgent, now, before it's too late |
| Tribal markers | we/they asymmetry, us vs them, real patriots |
| Dehumanizing terms | animals, vermin, horde, infestation |
| Attribution asymmetry | stated/confirmed vs claimed/alleged |
| Logical fallacies | whataboutism, false equivalence, ad hominem |
Step 3: Calculate Composite Factors
See references/scoring.md for weights.
COMPOSITE FACTORS:
─────────────────
Emotional Manipulation: [weighted avg of cat 1-5] → [1-5 scale]
Suspicious Timing: [weighted avg of cat 6-8] → [1-5 scale]
Uniform Messaging: [weighted avg of cat 9-11] → [1-5 scale]
Tribal Division: [weighted avg of cat 12-14] → [1-5 scale]
Missing Information: [weighted avg of cat 15-20] → [1-5 scale]
Step 4: Calculate Overall Score
OVERALL SCORE = Σ(composite_factor × weight × confidence)
Weights:
- Emotional Manipulation: 25%
- Suspicious Timing: 20%
- Uniform Messaging: 20%
- Tribal Division: 15%
- Missing Information: 20%
Scale 1-5 → 0-100: overall_score = (weighted_avg - 1) × 25
Step 5: Check Deep Research Triggers
After calculating scores, check if deep research is needed for claim verification.
Trigger Conditions (if ANY are met, proceed to verification):
DEEP RESEARCH CHECK:
─────────────────────
Overall NCI Score > 40? [ ] Yes → Verify key claims
Suspicious Timing > 3? [ ] Yes → Correlate events, timeline
Authority Issues (Cat 16) > 3? [ ] Yes → Verify credentials
Cherry-Picking (Cat 18) > 3? [ ] Yes → Find omitted context
Historical Parallels > 2? [ ] Yes → Research precedent campaigns
TRIGGERS MET: [N] → If > 0, proceed to verification
If Triggers Met:
Extract Key Claims: Identify 3-5 most impactful factual assertions
Invoke Claim Verifier: Use
fact-checkeragent or/decipon:verifyApply Deep Research: Use
../deep-research/SKILL.mdmethodologyTrack Results:
CLAIM: [Statement] STATUS: [VERIFIED / PARTIALLY VERIFIED / UNVERIFIED / CONTRADICTED] SOURCE: [URL] CONFIDENCE: [1-100] NCI IMPACT: [How this affects scores]Adjust Scores If Needed:
- Verified claims → May reduce Authority Issues, Cherry-Picking scores
- Contradicted claims → Increase relevant category scores
- Document adjustments in final report
If No Triggers Met: Proceed directly to Step 6 (Perspective Generation).
Step 6: Generate Dual Perspectives
CRITICAL: Always generate BOTH interpretations.
MANIPULATIVE INTERPRETATION:
This content appears designed to [specific manipulation goal].
Key manipulation techniques detected:
- [Technique 1 with evidence]
- [Technique 2 with evidence]
- [Technique 3 with evidence]
Confidence: [X]%
LEGITIMATE INTERPRETATION:
This content may reflect [genuine intent/concern].
Supporting factors:
- [Factor 1]
- [Factor 2]
- [Factor 3]
Confidence: [Y]%
For perspective generation guidance, leverage the critique framework from the deep-research skill if available.
Step 7: Output Report
Standard Format (Markdown):
# NCI Analysis Report
## Content Summary
[Brief description of analyzed content]
## Overall Score: [0-100] [severity indicator]
Confidence: [X]%
## Composite Factors
| Factor | Score | Confidence |
|--------|-------|------------|
| Emotional Manipulation | [X.X]/5 | [%] |
| Suspicious Timing | [X.X]/5 | [%] |
| Uniform Messaging | [X.X]/5 | [%] |
| Tribal Division | [X.X]/5 | [%] |
| Missing Information | [X.X]/5 | [%] |
## Key Findings
[Top 3-5 manipulation indicators with evidence]
## Claim Verification (if deep research triggered)
| Claim | Status | Confidence | Source |
|-------|--------|------------|--------|
| [Claim 1] | [VERIFIED/etc] | [%] | [URL] |
| [Claim 2] | [Status] | [%] | [URL] |
**Score Adjustment**: [Original] → [Adjusted] ([+/-N] due to verification)
## Perspectives
### If Manipulative
[Manipulative interpretation]
### If Legitimate
[Legitimate interpretation]
## Category Details
[Expandable section with all 20 category scores]
## Methodology
NCI Protocol v1.0 - Pattern-based manipulation detection
Deep Research: [Yes/No] - [N] claims verified
Severity Indicators (NCI Protocol v1.0):
- 0-25:
[·]Low manipulation risk - 26-50:
[!]Moderate - some concerning patterns - 51-75:
[!!]High - strong manipulation patterns - 76-100:
[!!!]Severe - overwhelming manipulation signs
Integration with Deep Research
This plugin includes the deep-research skill for fact-checking and claim verification. Reference: ../deep-research/SKILL.md
Automatic Triggers
Deep research is recommended when NCI analysis shows:
| Trigger | Threshold | Verification Focus |
|---|---|---|
| Overall NCI Score | > 40 (upper Moderate) | Verify key claims |
| Suspicious Timing | > 3 | Correlate events, check timeline |
| Authority Issues | > 3 | Verify credentials, expertise claims |
| Cherry-Picking | > 3 | Find omitted context, full data |
| Historical Parallels | > 2 | Research precedent campaigns |
Workflow Integration
NCI + DEEP RESEARCH WORKFLOW:
─────────────────────────────
1. Complete NCI analysis (Steps 1-6)
2. Check trigger conditions
3. If triggered:
- Extract key factual claims
- Apply claim-verifier agent
- Use deep research methodology
- Update scores based on findings
4. Generate final report with verification status
Using the Claim Verifier
After NCI analysis, invoke the claim-verifier agent:
- See
../agents/claim-verifier.mdfor verification workflow - Uses source evaluation from
../deep-research/references/source-evaluation.md - Applies critique framework from
../deep-research/references/critique-framework.md
Verification Commands
| Command | Purpose |
|---|---|
/decipon:analyze |
Pattern analysis (this skill) |
/decipon:verify |
Fact-check claims with deep research |
/decipon:report |
Combined analysis + verification report |
Source Evaluation Integration
When assessing sources during NCI analysis, apply confidence scoring:
| Source Type | Confidence | NCI Consideration |
|---|---|---|
| Official documentation | 85-95 | Reduces Authority Issues if verified |
| Government/institutional | 75-90 | Check for political context |
| Major news (AP, Reuters) | 70-85 | Generally reliable baseline |
| Partisan outlets | 40-60 | Note bias, affects Tribal Division |
| Anonymous/undated | 10-30 | Increases Missing Information |
See ../deep-research/references/source-evaluation.md for detailed scoring.
Contradiction Handling
When sources disagree during verification:
- Note the contradiction explicitly
- Apply confidence scoring to each source
- Research additional sources to resolve
- If unresolved, present both perspectives in report
See ../deep-research/references/critique-framework.md for resolution protocol.
Examples
See references/examples.md for historical case studies including:
- Nayirah Testimony (1990) - Score: 88
- Tobacco Industry Campaign - Score: 82
- Modern examples with full category breakdowns
When to Use
Use NCI Analysis:
- Content claiming urgent action needed
- Viral stories with strong emotional triggers
- Content creating clear us-vs-them dynamics
- Stories suspiciously timed with political events
- Content from unknown or questionable sources
Don't Use:
- Simple factual lookups (use fact-checking)
- Opinion pieces clearly labeled as such
- Personal correspondence
- Fiction/entertainment
References
- references/categories.md - All 20 category definitions with detection signals
- references/scoring.md - Scoring methodology, weights, and quantitative formulas
- references/examples.md - Historical case studies for calibration
- references/vocabulary.md - Detection vocabulary lists
- references/guidance.md - Actionable tips per factor