Overview

/check is a production command in the Quality category of the Prismatic Platform that provides comprehensive verification and integrity checking across session claims, commit messages, report metrics, and platform state. The command ensures that every claim made during development sessions is backed by evidence, every commit accurately describes its changes, and every metric reported in documentation matches the actual platform state.

In a complex platform with hundreds of agents, thousands of files, and continuous evolution, drift between claimed state and actual state is inevitable without systematic verification. The /check command serves as the platform's truth arbiter, cross-referencing assertions against live evidence at three configurable depth levels: quick (syntax and existence), thorough (full cross-reference with tests), and forensic (deep historical analysis with reproducibility verification).

This command operates under the L2+ authority level and is executed by the verification-integrity-commander agent. It is part of the platform's 216-command slash command registry, built on the AIAD (Autonomous Intelligence Agent Design) standard. The command supports five verification scopes (session, commit, reports, claims, all) and produces structured verification reports with confidence scores and recommended actions.

The verification philosophy aligns with the NABLA framework's Provenance Mandatory axiom: all beliefs about platform state must be traceable to verifiable evidence. The /check command operationalizes this axiom by systematically validating that documented claims have concrete backing in the codebase, test results, and compilation output.

Architecture

The verification system uses a multi-pass validation architecture where each scope runs independent verification pipelines that converge into a unified integrity report.

User Request (/check [scope] [depth])
    |
    v
Scope Router ──> Session | Commit | Reports | Claims | All
    |
    v
Verification Pipeline (per scope)
    |-- Evidence Collection
    |-- Cross-Reference Analysis
    |-- Discrepancy Detection
    |-- Confidence Scoring
    |
    v
Unified Integrity Report
    |-- VERIFIED / PARTIAL / FAILED
    |-- Discrepancy Details
    |-- Recommended Actions

Verification Scopes

Scope	What It Verifies	Evidence Sources
session	Session context claims, completion status	`.claude/session-context/`, git log
commit	Commit messages vs actual changes	git diff, compilation, tests
reports	Report metrics and statistics	Mix tasks, file counts, test results
claims	Arbitrary claims about platform state	Codebase analysis, runtime checks
all	Comprehensive platform-wide verification	All sources combined

Usage

Basic Verification

# Verify current session claims (default)
/check session

# Quick sanity check across all scopes
/check all quick

# Deep commit verification
/check commit thorough

# Full platform forensic check
/check all forensic

Scope-Specific Verification

# Verify session context accuracy
/check session thorough

# Validate report metrics against reality
/check reports forensic

# Cross-reference all claims in documentation
/check claims thorough

# Quick compilation and test sanity check
/check quick

Advanced Verification

# Verify specific session file
/check session --file 2025-12-13-archer-supreme-session.md

# Verify last N commits
/check commit --count 5 --depth thorough

# Verify claims in specific document
/check claims --file CLAUDE.md --depth forensic

Options & Parameters

Parameter	Type	Default	Description
`scope`	string	`session`	Verification scope: session, commit, reports, claims, all
`verification-type`	string	`thorough`	Depth: quick, thorough, forensic
`--file`	string	none	Specific file to verify
`--count`	integer	1	Number of commits to verify
`--output`	string	`terminal`	Output: terminal, file, json

Verification Levels

Level	Duration	Coverage	Best For
quick	< 30 seconds	Syntax, existence, basic metrics	Pre-commit sanity check
thorough	1-5 minutes	Full cross-reference, tests, analysis	Standard verification
forensic	5-30 minutes	Deep analysis, history, reproducibility	Release validation

Quick Verification Checks

File existence for all referenced paths
Syntax validity of configuration files
Basic metric ranges (file counts, agent counts)
Compilation status (warnings-as-errors)

Thorough Verification Checks

All quick checks plus:
Full cross-reference validation across documents
Test suite execution and result verification
Git log analysis against claimed changes
Quality gate execution and scoring
Agent registry integrity

Forensic Verification Checks

All thorough checks plus:
Historical claim tracking across sessions
Reproducibility testing of reported metrics
Statistical anomaly detection in reported numbers
Source code analysis for undocumented changes
Dependency audit for security vulnerabilities

Execution Flow

PHASE 1: SCOPE SELECTION
    |-- Parse scope and verification level
    |-- Load relevant evidence sources
    |-- Initialize verification pipeline
    |
PHASE 2: EVIDENCE COLLECTION
    |-- Gather claims from target scope
    |-- Collect supporting evidence
    |-- Load baseline metrics
    |
PHASE 3: CROSS-REFERENCE ANALYSIS
    |-- Match claims against evidence
    |-- Identify unsupported claims
    |-- Detect contradictions
    |-- Calculate confidence scores
    |
PHASE 4: DISCREPANCY DETECTION
    |-- Flag claims without evidence
    |-- Identify metric mismatches
    |-- Detect stale references
    |-- Classify severity of discrepancies
    |
PHASE 5: REPORT GENERATION
    |-- Compile verification results
    |-- Calculate overall integrity score
    |-- Generate recommended actions
    |-- Assign verification status
    |
PHASE 6: OUTPUT
    |-- Display formatted report
    |-- Save to verification history
    |-- Emit telemetry events

Output Format

Verification Report Structure

VERIFICATION REPORT
Status: VERIFIED | PARTIAL | FAILED

Claims Verified: 47/50 (94%)
Discrepancies Found: 3
Confidence Score: 0.94

DISCREPANCIES:
1. [MINOR] Agent count claimed 258, actual 255
   Source: CLAUDE.md line 42
   Evidence: mix git_trees find ".agent.md" | wc -l

2. [WARNING] Session file references non-existent path
   Source: session-context/2025-12-13-session.md
   Evidence: File not found at referenced path

3. [INFO] Quality score not verified (mix quality.gates not run)
   Action: Run mix quality.gates to verify

RECOMMENDED ACTIONS:
- Update CLAUDE.md agent count to 255
- Remove stale file reference from session context
- Run quality gates for score verification

Status Levels

Status	Meaning	Action Required
VERIFIED	All claims confirmed with evidence	None
PARTIAL	Some discrepancies found	Review and correct
FAILED	Significant integrity issues detected	Immediate investigation

Integration Points

Component	Relationship	Details
Prismatic Agents	Executed by `verification-integrity-commander`	Primary verification agent
AIAD Registry	Command specification and discovery	Standard AIAD interface
Quality Gates	Direct quality gate enforcement	Gate results as evidence
Telemetry	Command execution metrics	Verification event tracking
Git Integration	Commit and history analysis	Evidence for commit scope
Session Context	Session claim verification	`.claude/session-context/` files
Mix Tasks	Metric verification	`mix compile`, `mix test`, `mix quality.gates`

Best Practices

Run After Every Session: Execute /check session at the end of each development session to ensure session context accurately reflects what was accomplished.
Verify Before Deployment: Use /check all forensic before production deployments to ensure all platform claims are accurate and no undocumented changes exist.
Automate in Pre-Commit: Add /check commit quick to pre-commit hooks for automated commit message validation.
Cross-Reference Reports: When writing session reports or documentation updates, run /check reports thorough to catch metric inaccuracies before they propagate.
Track Verification History: Maintain verification reports in .claude/reports/verification/ for audit trail purposes and trend analysis.

Error Handling

Error	Cause	Resolution
`EVIDENCE_UNAVAILABLE`	Cannot access evidence source	Ensure mix tasks are available, git is accessible
`COMPILATION_REQUIRED`	Source needs compilation for verification	Run `mix compile` first
`TIMEOUT`	Forensic verification exceeded time limit	Use thorough level or narrow scope
`SCOPE_INVALID`	Unrecognized verification scope	Use: session, commit, reports, claims, all

Advanced Usage

Automated Verification Pipeline

# Pre-release verification chain
/check all forensic
/check commit --count 20 --depth thorough
/check reports forensic --output json > verification-report.json

Integration with Workflows

The /check command integrates with the AIAD workflow system:

Post-evolution: Verify evolution claims after /evolve execution
Pre-commit: Validate commit accuracy before staging
Release gate: Forensic verification as deployment prerequisite

Performance

Metric	Value	Notes
Quick Check	< 30 seconds	Syntax and existence only
Thorough Check	1-5 minutes	Full cross-reference
Forensic Check	5-30 minutes	Deep analysis
Per-Claim Verification	~2 seconds	Individual claim checking

Doctrine Compliance

All commands operate under the NO MERCY, NO DOUBTS doctrine:

NO MERCY: Zero tolerance for unverified claims. Every assertion about platform state must be backed by concrete evidence. Discrepancies are flagged immediately.
NO DOUBTS: Full investigation before action, evidence-based results. The verification system provides complete provenance for every claim validation.

The command directly implements the NABLA framework's Provenance Mandatory axiom: all beliefs must be traceable to verifiable evidence sources.

/quality-gates - Enforce quality gate checkpoints with zero-warning compilation validation
/quality-enforce - Mandatory progressive quality debt elimination with AIAD enforcement
/regression-check - Execute 25 custom Credo regression checks preventing 700+ violations
/code - Core coding implementation and feature development
/fix - Bug fix implementation with mandatory regression tests
/refactor - Safe refactoring with zero-regression guarantee
/debrief - Comprehensive session debrief with platform state analysis

Connect & Contribute

Created by Tomáš Korcak (korczis) | Open Source under GHL

GitHub | GitLab | LinkedIn | Contact
Developer Portal | Architecture | Meet the Creator

/check