Work Verification Skill
Critical Requirement: This skill forces the AI agent to rigorously verify every claim, statement, and output. No assumptions. No skipping. Everything must be double-checked with concrete evidence.
Applies to: Code, writing, analysis, research, documentation, data interpretation, summaries, reports, recommendations, and all other work products.
Usage Modes
Important: When the user invokes this skill, check what they want verified:
- If they mention file paths → verify those files
- If they mention specific claims → verify those claims
- If they say just
/verify→ verify recent session work - If unclear → ask them what they want verified
Mode 1: Verify Recent Session Work
When invoked without arguments (/verify), verify all work produced in the current session.
Example: /verify
Mode 2: Verify Specific Files
When the user provides file path(s), verify all claims in those specific files:
/verify path/to/report.md/verify analysis.txt data-summary.md/verify reports/*.md- User message: "Can you verify the Q4 report?" → Read the file and verify it
This mode is useful for:
- Fact-checking existing documents
- Verifying reports written outside this session
- Auditing analysis done by others
- Checking legacy documentation
- Reviewing work done by colleagues
Mode 3: Verify Specific Claims
When the user provides specific claims to check:
/verify "User engagement increased 25%"/verify the statistics in section 3- User message: "Check if the revenue numbers are accurate" → Find and verify revenue numbers
Verification Protocol
You MUST follow this protocol:
1. Identify All Claims
Go through everything you produced and list every factual claim or assertion you made. Include:
For Code Work:
- Statements about what code does
- Claims about file locations or contents
- Assertions about how systems work
- Descriptions of changes made
- Explanations of behavior
- Documentation of features
For Analytical & Writing Work:
- Statistical claims or data points
- Quotes or paraphrases from sources
- Numerical figures or calculations
- Date/time references
- Citations or attributions
- Causal relationships ("X causes Y")
- Comparisons ("A is larger than B")
- Historical facts or events
- Definitions or explanations
- Summary statements
- Conclusions drawn from data
2. Verify Each Claim
For EVERY claim identified above:
State the claim explicitly
Determine verification method: What evidence would prove or disprove this?
Gather evidence: Use tools to get concrete proof
For Code/Technical Work:
- Read actual files (don't rely on memory)
- Run actual commands (don't assume output)
- Check actual git history (don't guess)
- Search actual code (don't presume)
For Analytical/Writing Work:
- Re-read source documents/files
- Re-calculate numbers and statistics
- Check original sources for quotes
- Verify dates and times mentioned
- Cross-reference claims across multiple sources
- Search for contradicting information
- Validate logical inferences
- Check data sources and methodology
Compare: Does the evidence match the claim?
Document result:
- ✅ VERIFIED: Evidence confirms claim
- ❌ INCORRECT: Evidence contradicts claim
- ⚠️ PARTIAL: Evidence partially supports claim
- ❓ UNVERIFIABLE: Cannot gather evidence (state why)
3. Output Verification Table
At the end of your verification, provide a table with this format:
| # | Claim | Verification Method | Evidence | Result |
|---|---|---|---|---|
| 1 | "The function foo() is in src/main.py" |
Read file and search for function | File read + grep for "def foo" at line 42 | ✅ VERIFIED |
| 2 | "The config uses port 8080" | Read config file | Found port: 3000 in config.yml line 15 |
❌ INCORRECT |
| 3 | "User engagement increased 25%" | Re-read report + recalculate | Source shows 23% increase (from 1000 to 1230) | ❌ INCORRECT |
| 4 | "Document states 'high priority'" | Search source document | Exact quote found in section 2.3 | ✅ VERIFIED |
| 5 | "Total budget is $50,000" | Sum all line items | Actual total: $48,750 (calculation error) | ❌ INCORRECT |
| 6 | "Policy implemented Q3 2024" | Read meeting notes from Sept 2024 | Implementation date: July 15, 2024 (Q3) | ✅ VERIFIED |
4. Corrections
If any claims were incorrect or partially correct:
- Acknowledge the error: Be explicit about what was wrong
- Provide correct information: Based on actual evidence
- Explain the discrepancy: If known, why was the original claim wrong?
- Update any affected work: Fix documentation, code, or explanations
Verification Examples
Code Work Examples
Example 1: Code Change Verification
Claim: "I added error handling to the process_data() function"
Verification:
- Read the file containing
process_data() - Search for try/catch or error handling patterns
- Compare with git diff to see what was actually added
Example 2: File Location Verification
Claim: "The configuration is stored in config/settings.json"
Verification:
- Use Glob to find all config files
- Read the file at the claimed path
- Verify it contains configuration data
Example 3: Behavior Explanation Verification
Claim: "The system retries failed requests 3 times"
Verification:
- Find the retry logic in code
- Read the actual implementation
- Check for the retry count constant or configuration
Analytical & Writing Work Examples
Example 4: Statistical Claim Verification
Claim: "The report shows a 25% increase in user engagement"
Verification:
- Re-read the source report/data file
- Locate the actual engagement metrics
- Recalculate the percentage change
- Verify the baseline and comparison periods
Example 5: Quote Verification
Claim: "According to the document, 'performance improved significantly'"
Verification:
- Read the source document
- Search for the exact quote
- Verify surrounding context
- Check if paraphrasing is accurate
Example 6: Date/Historical Fact Verification
Claim: "The policy was implemented in Q3 2024"
Verification:
- Read meeting notes or policy documents
- Search for implementation dates
- Verify against timeline documents
- Cross-reference with related events
Example 7: Calculation Verification
Claim: "The total cost is $15,420 across 3 departments"
Verification:
- Re-read the source data
- Manually recalculate the sum
- Verify all departments are included
- Check for any rounding or currency conversions
Example 8: Summary Verification
Claim: "The main findings are X, Y, and Z"
Verification:
- Re-read the entire source material
- Identify all key findings mentioned
- Verify no major findings were omitted
- Check if X, Y, Z accurately reflect priorities
Example 9: Causal Relationship Verification
Claim: "The new feature caused user retention to increase"
Verification:
- Read the analysis methodology
- Check if other factors were controlled for
- Verify timeline (correlation vs causation)
- Look for conflicting evidence or alternative explanations
Important Rules
- Never skip verification: Every single claim must be verified
- Don't rely on memory: Always read/check actual sources
- Document uncertainty: If you can't verify, say so explicitly
- Be honest about errors: If you were wrong, admit it clearly
- Verify verifications: Double-check your verification methods are sound
- Update as you verify: If you find errors, fix them immediately
Common Verification Pitfalls
General:
- ❌ "I'm pretty sure this is correct" → ✅ Verify with evidence
- ❌ "This should be right" → ✅ Check the actual source
- ❌ "I recall that..." → ✅ Re-read the original material
- ❌ "Approximately..." (when exact figures exist) → ✅ Get exact numbers
Code Work:
- ❌ "The file is probably in X directory" → ✅ Use Glob/Grep to confirm
- ❌ "Based on common patterns..." → ✅ Check this specific codebase
- ❌ "I added X feature" → ✅ Read the actual diff to confirm
- ❌ "This should work" → ✅ Test or read actual implementation
Analytical Work:
- ❌ "The document says something like..." → ✅ Find and quote exact text
- ❌ "Around 50 people..." → ✅ Get the exact number from source
- ❌ "This probably correlates with..." → ✅ Check if data actually shows correlation
- ❌ "The main point is..." → ✅ Re-read to verify it's actually the main point
- ❌ "According to studies..." → ✅ Cite the specific study/source
- ❌ "The data shows..." → ✅ Verify what the data actually shows (not interpretation)
When to Use This Skill
Use this skill:
- After making significant code changes
- After writing documentation or explanations
- After completing analytical work or research
- After writing reports, summaries, or recommendations
- After processing data or creating visualizations
- After making claims based on sources
- When the user asks you to verify your work
- Before marking tasks as complete
- When accuracy is critical
- Before presenting findings or conclusions
- After any work where facts, figures, or claims were stated
Workflow
For Session Work (Mode 1)
- Review your recent work (messages, code changes, files written, analysis, reports)
- Extract all factual claims (make a comprehensive list)
- Verify each claim systematically (use tools, gather evidence)
- Create the verification table (show all results transparently)
- Make corrections (fix any errors found)
- Summarize findings (how many verified, how many errors, what was fixed)
For Specific Files (Mode 2)
- Read the specified file(s) completely
- Extract all factual claims from the content (comprehensive list)
- For each claim, gather evidence from:
- Other files referenced in the document
- Source data mentioned
- Original sources for quotes
- Re-calculations for numbers
- Create the verification table showing what was checked
- Report findings: Which claims are accurate, which need correction
- Do NOT modify files unless explicitly asked - just report findings
For Specific Claims (Mode 3)
- Identify the specific claims the user wants verified
- Locate these claims in files or recent work
- Verify each claim with concrete evidence
- Create focused verification table for just those claims
- Report findings with details
Types of Work This Applies To
- Code & Development: Changes, features, bug fixes, architecture
- Documentation: Technical docs, user guides, README files
- Analysis: Data analysis, performance analysis, competitive analysis
- Research: Literature review, fact-checking, source verification
- Writing: Reports, articles, summaries, presentations
- Data Processing: Calculations, aggregations, transformations
- Recommendations: Decisions based on data or analysis
- Communication: Emails, messages with factual claims
Practical Usage Examples
Example: Verify an Existing Report
User: /verify reports/Q4-analysis.md
Agent should:
- Read
reports/Q4-analysis.mdcompletely - Extract all claims (statistics, dates, conclusions)
- Find source data files referenced
- Re-calculate any numbers mentioned
- Verify all quotes and attributions
- Create verification table
- Report: "Found 15 claims, 13 verified, 2 incorrect"
Example: Verify Specific Statistics
User: /verify the revenue numbers in financial-summary.txt
Agent should:
- Read
financial-summary.txt - Extract all revenue-related numbers
- Locate source data (spreadsheets, databases)
- Recalculate each figure
- Create verification table for revenue claims only
- Report findings
Example: Verify Multiple Documents
User: /verify docs/*.md
Agent should:
- Read all markdown files in docs/
- Extract claims from each file
- Verify systematically (may take multiple tool calls)
- Create comprehensive verification table
- Report findings per file
Example: Check Someone Else's Analysis
User: /verify analysis/competitor-research.md (written by colleague)
Agent should:
- Read the file without assuming it's correct
- Treat every statement as needing verification
- Find original sources mentioned
- Re-verify data and calculations
- Report findings objectively
- Do NOT modify the file unless asked
Converted and distributed by TomeVault — claim your Tome and manage your conversions.