code-ai-installer 3.0.0 → 4.0.1-a

package/domains/content/locales/en/.agents/skills/fact-checking/SKILL.md

@@ -1,231 +1,246 @@
----
-name: fact-checking
-description: Fact verification — typology, source hierarchy, severity, checking methods, PASS/FAIL threshold
----
-# Fact Checking
-
-## When to Use
-- At the **Reviewer** stage — before final approval of content containing data and claims.
-- When **creating data-driven content** — in parallel with `$data-storytelling`.
-- During **audit of existing content** — checking the relevance of old publications.
-- Whenever there is **any doubt** about the authenticity of a fact — it is better to check once.
-
-## Input
-
-| Field | Required | Description |
-|------|:-----------:|----------|
-| Text for checking | ✅ | Content containing factual claims |
-| Content Type | ✅ | Determines the level of strictness (see below) |
-| Author's Sources | ⬚ | If the author provided their sources — check them |
-| Result of `$source-verification` | ⬚ | If already passed — use it, do not duplicate |
-
-> If the text contains no factual claims (purely opinion, humor) — the skill **is not applied**. Mark as N/A.
-
-## Strictness Levels
-
-The strictness of the check depends on the content type and risks:
-
-| Level | Content Types | Requirements | Example |
-|---------|--------------|------------|--------|
-| 🔴 **Maximum** | Medical, financial, legal, scientific claims | Only peer-reviewed / official sources. 0 unverified. | Article about medications, investment review |
-| 🟡 **Standard** | Articles, case studies, analytics, email, longreads | Reliable sources (Tier 1-2). ≤ 1 unverified (Minor). | Blog post with data, carousel with statistics |
-| 🔵 **Basic** | Social media, short posts, entertaining content | Common sense + check key figures. Common knowledge facts without links are acceptable. | Meme with a number, post saying "according to research" |
-
-> The level determines how many checks are needed and what sources are acceptable.
-
-## Fact Typology
-
-| Fact Type | What to check | Checking Method | Example |
-|-----------|--------------|----------------|--------|
-| **Statistics / numbers** | Number, sample, year, methodology | Find original research | "87% of marketers use AI" |
-| **Quote** | Accuracy, context, authorship | Find the primary source of the quote | "Steve Jobs said: ..." |
-| **Date / chronology** | Correctness of the date, order of events | Several independent sources | "The company was founded in 2015" |
-| **Name / job title** | Correct spelling, current job title | Official website, LinkedIn | "Google CEO Sundar Pichai" |
-| **Scientific claim** | Scientific consensus, has it been refuted | Peer-reviewed journals, meta-analyses | "Coffee reduces the risk of diabetes" |
-| **Legal claim** | Current legislation, jurisdiction | Official legal databases | "By Russian law, advertising must be labeled" |
-| **Causal relationship** | Correlation ≠ causation | Check research methodology | "Content marketing increases sales by 30%" |
-| **Common knowledge** | Check if it is really common knowledge | Quick check if there are doubts | "The Earth revolves around the Sun" |
-
-## Source Hierarchy
-
-| Tier | Source Type | Reliability | Examples |
-|------|--------------|------------|---------|
-| **Tier 1** | Primary / academic | Maximum | Peer-reviewed journals, state statistics (Rosstat, BLS), SEC filings, patents |
-| **Tier 2** | Authoritative secondary | High | Industry reports (McKinsey, Gartner, Statista), major media (Reuters, Bloomberg), official company blogs |
-| **Tier 3** | Reliable secondary | Medium | Industry media, expert books, TED talks, major editorial blogs |
-| **Tier 4** | User-generated content | Low | Personal blogs, social media posts, Reddit, Quora, Wikipedia (without checking the primary source) |
-| **Tier 5** | Unreliable | Minimum | Anonymous sources, content farms, yellow press, affiliate sites |
-
-### Rules by Strictness Level
-- 🔴 **Maximum:** Tier 1-2 only. Tier 3 is allowed for context, but not for key claims.
-- 🟡 **Standard:** Tier 1-3. Tier 4 is allowed for illustrations, but not for numbers.
-- 🔵 **Basic:** Tier 1-4. Tier 5 is not allowed under any level.
-
-## Severity — Error Levels
-
-| Severity | Definition | Examples | Impact |
-|----------|-------------|---------|---------|
-| 🔴 **Critical** | Factual error that misleads, legal/medical risk | Incorrect medication dosage, false statistics, non-existent law | FAIL |
-| 🟡 **Major** | Noticeable inaccuracy that undermines trust | Outdated figure (2019 instead of 2025), wrong job title, distorted quote | FAIL if ≥ 2 |
-| 🔵 **Minor** | Minor inaccuracy not affecting the core message | Rounding (87% instead of 86.7%), inexact date (March instead of April) | Does not block |
-
-### PASS / FAIL Threshold
-
-| Result | Condition |
-|-----------|---------|
-| **VERIFIED** | 0 Critical + 0 Major + 0 unverified (🔴 strictness level) |
-| **PASS** | 0 Critical + ≤ 1 Major + ≤ 1 unverified Minor |
-| **PASS WITH NOTES** | 0 Critical + ≤ 1 Major (with correction) + ≤ 2 unverified Minor |
-| **FAIL** | ≥ 1 Critical OR ≥ 2 Major OR ≥ 3 unverified |
-
-## Verification Methods
-
-| Method | When to use | How |
-|-------|-------------------|-----|
-| **Primary source search** | Statistics, scientific claims | Find original study/report, not a summary |
-| **Cross-checking** | Dates, names, events | ≥ 2 independent sources match |
-| **Reverse quote search** | Quotes attributed to famous people | Web search exact quote + name → check attribution |
-| **Wayback Machine** | Data from deleted/modified pages | web.archive.org → snapshot on date |
-| **Methodology check** | Statistics with unusual conclusions | Who conducted? What sample? Who sponsored? |
-| **Relevance check** | Any data | Source publication date ≤ 2 years (for fast-paced markets ≤ 1 year) |
-| **Context check** | Quotes taken out of context | Read the original more broadly — 2-3 paragraphs around the quote |
-| **Calculator / common sense** | Numbers that "don't add up" | Recalculate: percentages, ratios, orders of magnitude |
-
-## Protocol
-
-### Step 0 — Preparation
-1. Receive text and determine the content type.
-2. Determine strictness level (🔴 / 🟡 / 🔵).
-3. Load `$source-verification` result, if available.
-
-### Step 1 — Fact Extraction
-1. Read the text entirely.
-2. Highlight **all** factual claims — each on a separate line.
-3. Classify each by typology (statistics / quote / date / name / scientific / legal / causation / common knowledge).
-4. Determine the **criticality** of each fact to the core of the content (key / supporting / illustrative).
-
-### Step 2 — Verification
-For each claim:
-1. Select verification method based on fact type.
-2. Find source (adhering to hierarchy by strictness level).
-3. Check accuracy: number, date, name, context.
-4. Check relevance: is the data outdated?
-5. Assign status: ✅ Correct / ❌ Incorrect / ⚠️ Unverified.
-
-### Step 3 — Error Evaluation
-For each ❌ and ⚠️:
-1. Assign severity (Critical / Major / Minor).
-2. For ❌ — indicate the correct version with a source.
-3. For ⚠️ — recommendation: find source / remove / replace with verified fact.
-
-### Step 4 — Verdict
-1. Count by severity.
-2. Apply PASS / FAIL threshold.
-3. Form a list of mandatory edits and recommended removals.
-
-## Validation (Quality Gate)
-
-Fact-checking is considered complete if:
-
-- [ ] All factual claims extracted from text
-- [ ] Each claim classified by type
-- [ ] Each claim verified (✅ / ❌ / ⚠️ — none empty)
-- [ ] For each ❌ specified: severity, correct version, source
-- [ ] For each ⚠️ specified recommendation (find / remove / replace)
-- [ ] Sources match strictness level
-- [ ] Verdict matches threshold (not subjective)
-- [ ] List of mandatory edits formed
-
-## Handoff
-
-| Result | Route |
-|-----------|---------|
-| **VERIFIED / PASS** | → `$content-review-checklist` (item #18 — link to this result) → `$content-release-gate` |
-| **PASS WITH NOTES** | → Copywriter (fix Major + replace/remove ⚠️) → repeated fact-checking |
-| **FAIL** | → Copywriter / Researcher (fix Critical/Major) → mandatory repeated fact-checking |
-
-When handing off — use `$handoff` attaching the full report.
-
-## Integration with other skills
-
-| Skill | Interaction |
-|------|----------------|
-| `$source-verification` | **Complement:** source verification → used as input for fact-checking |
-| `$data-storytelling` | **Source:** data from data story is checked by fact-checking |
-| `$topic-research` | **Source:** facts from topic research — check before use |
-| `$content-review-checklist` | **Consumer:** item #18 links to fact-checking result |
-| `$content-release-gate` | **Consumer:** VERIFIED/PASS is mandatory for Release Gate |
-
-## Anti-patterns
-
-| Mistake | Why it's bad | How to do it right |
-|--------|-------------|---------------|
-| Checking only "suspicious" facts | Obvious facts can also be wrong | Check all factual claims |
-| Wikipedia as sole source | Wikipedia is Tier 4, may contain errors | Wikipedia → find primary source from footnotes |
-| "According to research" without link | Unverifiable claim | Specific source: who, when, where, sample |
-| Old data without checking | "In 2020 the market was..." → what about 2026? | Check relevance, look for fresh data |
-| Correlation = causation | "Companies with blogs grow faster" ≠ blog → growth | Explicitly state: correlation, causation not proven |
-| Checking against summary instead of original | Summary might have distorted data | Always look for the primary source |
-| One source = verified | Source can be wrong | For key facts — ≥ 2 independent sources |
-| Fact is correct, but outdated | World has changed since publication | Relevance: ≤ 2 years, for fast markets ≤ 1 year |
-
-## Output Template
-
-```
-### Fact Checking — [Content Title]
-
-**Content Type:** [article / post / email / longread]
-**Strictness Level:** [🔴 Maximum / 🟡 Standard / 🔵 Basic]
-**Iteration:** [1 / 2 / N]
-
----
-
-#### Extracted Claims
-
-| # | Claim | Type | Criticality | Source | Tier | Status | Sev | Comment |
-|---|------------|-----|-------------|----------|------|--------|-----|-------------|
-| 1 | "[quote]" | Statistics | Key | [source, year] | Tier 1 | ✅ Correct | — | |
-| 2 | "[quote]" | Quote | Supporting | [source] | Tier 2 | ✅ Correct | — | Context verified |
-| 3 | "[quote]" | Statistics | Key | [source, year] | Tier 3 | ❌ Incorrect | 🟡 Major | Correct: [correct number + source] |
-| 4 | "[quote]" | Date | Illustrative | Not found | — | ⚠️ Unverified | 🔵 Minor | Recommendation: find source or remove |
-| 5 | "[quote]" | Causation | Key | [source] | Tier 2 | ⚠️ Correlation | 🟡 Major | Rephrase: correlation, not causation |
-
----
-
-#### Summary
-
-| Status | Amount | Items |
-|--------|-----------|--------|
-| ✅ Verified | [N] | #[list] |
-| ❌ Incorrect | [N] | #[list] |
-| ⚠️ Unverified | [N] | #[list] |
-
-| Severity | Amount | Items |
-|----------|-----------|--------|
-| 🔴 Critical | [N] | #[list] |
-| 🟡 Major | [N] | #[list] |
-| 🔵 Minor | [N] | #[list] |
-
----
-
-### Verdict: [ VERIFIED / PASS / PASS WITH NOTES / FAIL ]
-
-**Justification:** [by threshold rules]
-
----
-
-#### Mandatory Edits (❌)
-
-| # | Claim | Sev | Correct Version | Source |
-|---|------------|-----|--------------------|----------|
-| 3 | "[quote]" | 🟡 | [correct version] | [source, year, Tier] |
-| 5 | "[quote]" | 🟡 | Rephrase: "[variant]" | [source] |
-
-#### Recommended Actions (⚠️)
-
-| # | Claim | Sev | Recommendation |
-|---|------------|-----|-------------|
-| 4 | "[quote]" | 🔵 | Find source or remove |
-
-**→ Next step:** [→ $content-review-checklist (ref:$fact-checking) / → Copywriter with edits / → repeated fact-checking after fixes]
-```
+---
+name: fact-checking
+description: Fact verification — typology, source hierarchy, severity, checking methods, PASS/FAIL threshold
+type: triggered
+domain: content
+owners:
+  - copywriter
+gates:
+  - COPYWRITER
+tech: []
+topic: []
+triggers:
+  - "fact-checking"
+  - "Факт-чекинг"
+  - "Верификация фактов — типология, иерархия источников, severity, методы проверки, порог PASS/FAIL"
+related: []
+budget_lines: 246
+schema_version: 1
+---
+# Fact Checking
+
+## When to Use
+- At the **Reviewer** stage — before final approval of content containing data and claims.
+- When **creating data-driven content** — in parallel with `$data-storytelling`.
+- During **audit of existing content** — checking the relevance of old publications.
+- Whenever there is **any doubt** about the authenticity of a fact — it is better to check once.
+
+## Input
+
+| Field | Required | Description |
+|------|:-----------:|----------|
+| Text for checking | ✅ | Content containing factual claims |
+| Content Type | ✅ | Determines the level of strictness (see below) |
+| Author's Sources | ⬚ | If the author provided their sources — check them |
+| Result of `$source-verification` | ⬚ | If already passed — use it, do not duplicate |
+
+> If the text contains no factual claims (purely opinion, humor) — the skill **is not applied**. Mark as N/A.
+
+## Strictness Levels
+
+The strictness of the check depends on the content type and risks:
+
+| Level | Content Types | Requirements | Example |
+|---------|--------------|------------|--------|
+| 🔴 **Maximum** | Medical, financial, legal, scientific claims | Only peer-reviewed / official sources. 0 unverified. | Article about medications, investment review |
+| 🟡 **Standard** | Articles, case studies, analytics, email, longreads | Reliable sources (Tier 1-2). ≤ 1 unverified (Minor). | Blog post with data, carousel with statistics |
+| 🔵 **Basic** | Social media, short posts, entertaining content | Common sense + check key figures. Common knowledge facts without links are acceptable. | Meme with a number, post saying "according to research" |
+
+> The level determines how many checks are needed and what sources are acceptable.
+
+## Fact Typology
+
+| Fact Type | What to check | Checking Method | Example |
+|-----------|--------------|----------------|--------|
+| **Statistics / numbers** | Number, sample, year, methodology | Find original research | "87% of marketers use AI" |
+| **Quote** | Accuracy, context, authorship | Find the primary source of the quote | "Steve Jobs said: ..." |
+| **Date / chronology** | Correctness of the date, order of events | Several independent sources | "The company was founded in 2015" |
+| **Name / job title** | Correct spelling, current job title | Official website, LinkedIn | "Google CEO Sundar Pichai" |
+| **Scientific claim** | Scientific consensus, has it been refuted | Peer-reviewed journals, meta-analyses | "Coffee reduces the risk of diabetes" |
+| **Legal claim** | Current legislation, jurisdiction | Official legal databases | "By Russian law, advertising must be labeled" |
+| **Causal relationship** | Correlation ≠ causation | Check research methodology | "Content marketing increases sales by 30%" |
+| **Common knowledge** | Check if it is really common knowledge | Quick check if there are doubts | "The Earth revolves around the Sun" |
+
+## Source Hierarchy
+
+| Tier | Source Type | Reliability | Examples |
+|------|--------------|------------|---------|
+| **Tier 1** | Primary / academic | Maximum | Peer-reviewed journals, state statistics (Rosstat, BLS), SEC filings, patents |
+| **Tier 2** | Authoritative secondary | High | Industry reports (McKinsey, Gartner, Statista), major media (Reuters, Bloomberg), official company blogs |
+| **Tier 3** | Reliable secondary | Medium | Industry media, expert books, TED talks, major editorial blogs |
+| **Tier 4** | User-generated content | Low | Personal blogs, social media posts, Reddit, Quora, Wikipedia (without checking the primary source) |
+| **Tier 5** | Unreliable | Minimum | Anonymous sources, content farms, yellow press, affiliate sites |
+
+### Rules by Strictness Level
+- 🔴 **Maximum:** Tier 1-2 only. Tier 3 is allowed for context, but not for key claims.
+- 🟡 **Standard:** Tier 1-3. Tier 4 is allowed for illustrations, but not for numbers.
+- 🔵 **Basic:** Tier 1-4. Tier 5 is not allowed under any level.
+
+## Severity — Error Levels
+
+| Severity | Definition | Examples | Impact |
+|----------|-------------|---------|---------|
+| 🔴 **Critical** | Factual error that misleads, legal/medical risk | Incorrect medication dosage, false statistics, non-existent law | FAIL |
+| 🟡 **Major** | Noticeable inaccuracy that undermines trust | Outdated figure (2019 instead of 2025), wrong job title, distorted quote | FAIL if ≥ 2 |
+| 🔵 **Minor** | Minor inaccuracy not affecting the core message | Rounding (87% instead of 86.7%), inexact date (March instead of April) | Does not block |
+
+### PASS / FAIL Threshold
+
+| Result | Condition |
+|-----------|---------|
+| **VERIFIED** | 0 Critical + 0 Major + 0 unverified (🔴 strictness level) |
+| **PASS** | 0 Critical + ≤ 1 Major + ≤ 1 unverified Minor |
+| **PASS WITH NOTES** | 0 Critical + ≤ 1 Major (with correction) + ≤ 2 unverified Minor |
+| **FAIL** | ≥ 1 Critical OR ≥ 2 Major OR ≥ 3 unverified |
+
+## Verification Methods
+
+| Method | When to use | How |
+|-------|-------------------|-----|
+| **Primary source search** | Statistics, scientific claims | Find original study/report, not a summary |
+| **Cross-checking** | Dates, names, events | ≥ 2 independent sources match |
+| **Reverse quote search** | Quotes attributed to famous people | Web search exact quote + name → check attribution |
+| **Wayback Machine** | Data from deleted/modified pages | web.archive.org → snapshot on date |
+| **Methodology check** | Statistics with unusual conclusions | Who conducted? What sample? Who sponsored? |
+| **Relevance check** | Any data | Source publication date ≤ 2 years (for fast-paced markets ≤ 1 year) |
+| **Context check** | Quotes taken out of context | Read the original more broadly — 2-3 paragraphs around the quote |
+| **Calculator / common sense** | Numbers that "don't add up" | Recalculate: percentages, ratios, orders of magnitude |
+
+## Protocol
+
+### Step 0 — Preparation
+1. Receive text and determine the content type.
+2. Determine strictness level (🔴 / 🟡 / 🔵).
+3. Load `$source-verification` result, if available.
+
+### Step 1 — Fact Extraction
+1. Read the text entirely.
+2. Highlight **all** factual claims — each on a separate line.
+3. Classify each by typology (statistics / quote / date / name / scientific / legal / causation / common knowledge).
+4. Determine the **criticality** of each fact to the core of the content (key / supporting / illustrative).
+
+### Step 2 — Verification
+For each claim:
+1. Select verification method based on fact type.
+2. Find source (adhering to hierarchy by strictness level).
+3. Check accuracy: number, date, name, context.
+4. Check relevance: is the data outdated?
+5. Assign status: ✅ Correct / ❌ Incorrect / ⚠️ Unverified.
+
+### Step 3 — Error Evaluation
+For each ❌ and ⚠️:
+1. Assign severity (Critical / Major / Minor).
+2. For ❌ — indicate the correct version with a source.
+3. For ⚠️ — recommendation: find source / remove / replace with verified fact.
+
+### Step 4 — Verdict
+1. Count by severity.
+2. Apply PASS / FAIL threshold.
+3. Form a list of mandatory edits and recommended removals.
+
+## Validation (Quality Gate)
+
+Fact-checking is considered complete if:
+
+- [ ] All factual claims extracted from text
+- [ ] Each claim classified by type
+- [ ] Each claim verified (✅ / ❌ / ⚠️ — none empty)
+- [ ] For each ❌ specified: severity, correct version, source
+- [ ] For each ⚠️ specified recommendation (find / remove / replace)
+- [ ] Sources match strictness level
+- [ ] Verdict matches threshold (not subjective)
+- [ ] List of mandatory edits formed
+
+## Handoff
+
+| Result | Route |
+|-----------|---------|
+| **VERIFIED / PASS** | → `$content-review-checklist` (item #18 — link to this result) → `$content-release-gate` |
+| **PASS WITH NOTES** | → Copywriter (fix Major + replace/remove ⚠️) → repeated fact-checking |
+| **FAIL** | → Copywriter / Researcher (fix Critical/Major) → mandatory repeated fact-checking |
+
+When handing off — use `$handoff` attaching the full report.
+
+## Integration with other skills
+
+| Skill | Interaction |
+|------|----------------|
+| `$source-verification` | **Complement:** source verification → used as input for fact-checking |
+| `$data-storytelling` | **Source:** data from data story is checked by fact-checking |
+| `$topic-research` | **Source:** facts from topic research — check before use |
+| `$content-review-checklist` | **Consumer:** item #18 links to fact-checking result |
+| `$content-release-gate` | **Consumer:** VERIFIED/PASS is mandatory for Release Gate |
+
+## Anti-patterns
+
+| Mistake | Why it's bad | How to do it right |
+|--------|-------------|---------------|
+| Checking only "suspicious" facts | Obvious facts can also be wrong | Check all factual claims |
+| Wikipedia as sole source | Wikipedia is Tier 4, may contain errors | Wikipedia → find primary source from footnotes |
+| "According to research" without link | Unverifiable claim | Specific source: who, when, where, sample |
+| Old data without checking | "In 2020 the market was..." → what about 2026? | Check relevance, look for fresh data |
+| Correlation = causation | "Companies with blogs grow faster" ≠ blog → growth | Explicitly state: correlation, causation not proven |
+| Checking against summary instead of original | Summary might have distorted data | Always look for the primary source |
+| One source = verified | Source can be wrong | For key facts — ≥ 2 independent sources |
+| Fact is correct, but outdated | World has changed since publication | Relevance: ≤ 2 years, for fast markets ≤ 1 year |
+
+## Output Template
+
+```
+### Fact Checking — [Content Title]
+
+**Content Type:** [article / post / email / longread]
+**Strictness Level:** [🔴 Maximum / 🟡 Standard / 🔵 Basic]
+**Iteration:** [1 / 2 / N]
+
+---
+
+#### Extracted Claims
+
+| # | Claim | Type | Criticality | Source | Tier | Status | Sev | Comment |
+|---|------------|-----|-------------|----------|------|--------|-----|-------------|
+| 1 | "[quote]" | Statistics | Key | [source, year] | Tier 1 | ✅ Correct | — | |
+| 2 | "[quote]" | Quote | Supporting | [source] | Tier 2 | ✅ Correct | — | Context verified |
+| 3 | "[quote]" | Statistics | Key | [source, year] | Tier 3 | ❌ Incorrect | 🟡 Major | Correct: [correct number + source] |
+| 4 | "[quote]" | Date | Illustrative | Not found | — | ⚠️ Unverified | 🔵 Minor | Recommendation: find source or remove |
+| 5 | "[quote]" | Causation | Key | [source] | Tier 2 | ⚠️ Correlation | 🟡 Major | Rephrase: correlation, not causation |
+
+---
+
+#### Summary
+
+| Status | Amount | Items |
+|--------|-----------|--------|
+| ✅ Verified | [N] | #[list] |
+| ❌ Incorrect | [N] | #[list] |
+| ⚠️ Unverified | [N] | #[list] |
+
+| Severity | Amount | Items |
+|----------|-----------|--------|
+| 🔴 Critical | [N] | #[list] |
+| 🟡 Major | [N] | #[list] |
+| 🔵 Minor | [N] | #[list] |
+
+---
+
+### Verdict: [ VERIFIED / PASS / PASS WITH NOTES / FAIL ]
+
+**Justification:** [by threshold rules]
+
+---
+
+#### Mandatory Edits (❌)
+
+| # | Claim | Sev | Correct Version | Source |
+|---|------------|-----|--------------------|----------|
+| 3 | "[quote]" | 🟡 | [correct version] | [source, year, Tier] |
+| 5 | "[quote]" | 🟡 | Rephrase: "[variant]" | [source] |
+
+#### Recommended Actions (⚠️)
+
+| # | Claim | Sev | Recommendation |
+|---|------------|-----|-------------|
+| 4 | "[quote]" | 🔵 | Find source or remove |
+
+**→ Next step:** [→ $content-review-checklist (ref:$fact-checking) / → Copywriter with edits / → repeated fact-checking after fixes]
+```