sethmblack/paks-skills/analyze-references
Claude Code and agent skill by sethmblack/paks-skills
- community
Checking that the SKILL.md file is still in the repository.
Audit a references.md file to verify that all cited URLs are real, accessible, and actually support the claims attributed to them.
Install
npx vskill@latest install sethmblack/paks-skills/analyze-referencesSKILL.md
The same SKILL.md was published earlier as sethmblack/paks-skills/analyze-references.
Analyze References
Audit a references.md file to verify that all cited URLs are real, accessible, and actually support the claims attributed to them.
Token Budget: ~400 tokens Purpose: Quality control for fact-checking; catch hallucinated sources
When to Use
- After creating a references.md file
- Auditing existing expert persona references
- "Verify these sources are real"
- "Check for hallucinated URLs"
- Quality control before publishing
The Problem
LLMs can hallucinate plausible-looking URLs and citations that:
- Don't exist at all (404)
- Exist but don't contain the claimed information
- Are real sites but wrong pages
- Have correct domains but fabricated paths
This skill systematically verifies each reference is real and accurate.
Inputs
| Input | Required | Description | |
|---|---|---|---|
| references_file | Yes | Path to the references.md file to audit | |
| thoroughness | No | "quick" (spot-check) or "full" (every URL) |
Audit Protocol
Step 1: Extract All URLs
Parse the references.md file and extract every URL cited. Group by category:
- Wikipedia links
- Encyclopedia/reference sites (Britannica, etc.)
- Official sources (company sites, .gov, etc.)
- News articles
- Books/publications
- Other
Step 2: Verify URL Existence
For each URL:
- Fetch the URL using WebFetch
- Check for 404/errors - Does the page exist?
- If redirected - Note the redirect destination
Step 3: Verify Content Match
For URLs that exist:
- Search for key terms from the claim on the page
- Confirm the page actually supports the cited fact
- Flag mismatches where URL exists but doesn't contain claimed info
Step 4: Categorize Results
| Status | Meaning | Action | |
|---|---|---|---|
| ✓ VERIFIED | URL exists and supports claim | None | |
| ⚠ REDIRECT | URL redirects elsewhere | Update URL | |
| ⚠ PARTIAL | Page exists but claim not found verbatim | Review manually | |
| ✗ BROKEN | 404 or page doesn't exist | Find replacement or remove | |
| ✗ MISMATCH | Page exists but contradicts claim | Correct the claim |
Workflow
Step 1: Gather and Review Inputs
Collect all relevant information:
- Review the provided data and context
- Identify key parameters and constraints
- Clarify any ambiguities or missing information
- Establish success criteria
Step 2: Analyze the Situation
Perform systematic analysis:
- Identify patterns and relationships
- Evaluate against established frameworks
- Consider multiple perspectives
- Document key findings
Step 3: Generate Recommendations
Create actionable outputs:
- Synthesize insights from analysis
- Prioritize recommendations by impact
- Ensure recommendations are specific and measurable
- Consider implementation feasibility
Output Format
Common Hallucination Patterns
Watch for these red flags:
| Pattern | Example | Why Suspicious | |
|---|---|---|---|
| Too-specific paths | /article/2024/specific-fact-123 | LLMs invent plausible paths | |
| Round dates in URLs | /2020/01/01/ | Suspiciously clean dates | |
| Quote in URL | britannica.com/quote/exact-words | Quotes aren't typically in URLs | |
| Mixed domain styles | en.wiki.org vs en.wikipedia.org | Slight domain errors | |
| Archived versions | web.archive.org/... | May not have actually checked |
Quick Audit (Spot-Check)
For "quick" thoroughness:
- Check all Wikipedia URLs (easy to verify)
- Check any URL that looks suspiciously specific
- Spot-check 20% of remaining URLs randomly
- Check all URLs for claims marked as "key facts"
Outputs
Primary Output: A structured analysis document that identifies and articulates patterns, insights, and actionable recommendations based on the input data.
Format:
Example output: See the Example section below for a complete demonstration.
Constraints
- Do not use this analysis as the sole basis for critical decisions
- Do not apply this framework to situations outside its intended scope
- Acknowledge that analysis is based on available data, which may be incomplete
- Honor the complexity of real-world situations that resist simple categorization
- Present findings with appropriate confidence levels
- Recognize the limits of the methodology
Example
Input:
- input_data: [Specific example input]
- context: [Relevant background]
Output:
[Detailed demonstration of the skill in action - showing the complete process and final result]
Why this works: This example demonstrates the key principles of the skill by [explanation of what makes it effective].
Integration
This skill is part of the expert creation workflow. Run after:
fact-check-expertcreates references.md- Before finalizing any expert persona
Can be invoked standalone for auditing existing references.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckinganalyze-references/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/sethmblack/paks-skills/analyze-references)