verify-changes — Claude Code command: what it does, how to install
verify-changes executes adversarial multi-subagent verification across recent code diffs to weed out bugs and false alarms.
MIT ★ 1,424 updated 2026-10-06
.claude/commands/verify-changes.mdWhat verify-changes does
The command inspects the latest repository changes via git diff HEAD~1 and launches five background subagents in a single message. These subagents simultaneously validate syntax with tools like tsc or mypy, execute targeted tests, inspect style via linters, compile build targets, and grep for leaked credentials.
Raw findings do not go straight into the report. A secondary trio of adversarial subagents reviews every flagged item to filter false positives, uncover missing edge cases, and ensure reported issues are not intentional patterns. The final summary presents pass/fail verdicts alongside confirmed flaws.
Parallel execution relies strictly on the Task tool supporting run_in_background: true in your environment.
What it helps with
- verify recent commits across typing, test suites, builds, and linters
- filter out false positives from automated verification passes
- scan modified files for committed tokens and credentials
- catch missed edge cases and faulty error handling after edits
Example prompts for the agent
- /project-starter:verify-changes
- Run /project-starter:verify-changes to inspect the latest diff
- Verify my recent edits using /project-starter:verify-changes
How to install verify-changes
Manually: put the files in place
.claude/commands/verify-changes.mdCall it with /verify-changes. For all projects use ~/.claude/commands/
Manually: put the files in place
~/.codex/prompts/verify-changes.mdCall it with /prompts:verify-changes. Codex marks this format deprecated in favour of skills
Manually: put the files in place
.github/prompts/verify-changes.prompt.mdManually: put the files in place
.gemini/commands/verify-changes.tomlGemini stores commands as TOML: the prompt text goes into the prompt field
Questions about verify-changes
How to install verify-changes in Claude Code?
Place the command file at .claude/commands/verify-changes.md.
What does verify-changes do?
It reads changed files from Git, runs five automated verification subagents simultaneously, and refines the findings using three adversarial reviewers.
Why does it use an adversarial review phase?
Three specialized agents review the initial findings to discard non-issues, check project context, and catch edge cases missed during the first pass.
Source text
verify-changes.md · MIT
---
description: Comprehensive verification after code changes. Uses Boris Cherny's multi-subagent adversarial approach.
---
# Verify Changes
Run comprehensive verification of recent code changes using multi-subagent adversarial approach.
## Phase 1: Gather Context
Run these commands to understand what changed:
```bash
git diff HEAD~1 --stat
git log -3 --oneline
git diff --name-only HEAD~1
```
Identify affected files and their corresponding test files.
## Phase 2: Spawn Verification Subagents (Parallel)
Use the Task tool with `run_in_background: true` to spawn subagents in parallel.
**CRITICAL**: Launch ALL subagents in a SINGLE message for true parallelism:
```
[All Task calls in ONE message]
Task 1: {
description: "Syntax & Type Check",
prompt: "Run language-specific type checker (tsc, mypy, go vet)...",
run_in_background: true
}
Task 2: {
description: "Test Runner",
prompt: "Run tests for affected files first, then integration tests...",
run_in_background: true
}
Task 3: {
description: "Lint & Style Check",
prompt: "Run project linter (eslint, ruff, golangci-lint)...",
run_in_background: true
}
Task 4: {
description: "Security Scan",
prompt: "Grep for hardcoded secrets, check for vulnerabilities...",
run_in_background: true
}
Task 5: {
description: "Build Validator",
prompt: "Run build command, verify artifacts exist...",
run_in_background: true
}
```
Each subagent returns its result automatically when it completes — there is no
separate retrieval call. Read each verifier's report as it returns.
### Subagent Details
### Subagent 1: Syntax & Type Check
- Run language-specific type checker (tsc, mypy, go vet)
- Report any type errors or warnings
- Exit with list of issues found
### Subagent 2: Test Runner
- Run tests for affected files first
- Run related integration tests
- Report failures with file:line context
### Subagent 3: Lint & Style Check
- Run project linter (eslint, ruff, golangci-lint)
- Check for formatting issues
- Report violations with severity
### Subagent 4: Security Scan
- Grep for hardcoded secrets (passwords, API keys, tokens)
- Check for common vulnerabilities in new code
- Review any new dependencies
### Subagent 5: Build Validator
- Run build command
- Verify build artifacts exist
- Check for build warnings
## Phase 3: Adversarial Review
Spawn 3 adversarial subagents to review Phase 2 findings:
### Adversarial A: False Positive Filter
- Review each finding from Phase 2
- Determine if it's a real issue or false alarm
- Provide reasoning for each determination
### Adversarial B: Missing Issues Finder
- Look for issues the first pass might have missed
- Check edge cases in changed code
- Verify error handling is adequate
### Adversarial C: Context Validator
- Verify findings make sense in project context
- Check if "issues" are actually intentional patterns
- Consider project conventions
## Phase 4: Synthesize Results
Combine all subagent outputs into final report:
```
## Verification Results: [PASS/FAIL]
### Confirmed Issues
1. [Issue] - [Location] - [Why it's confirmed real]
### Warnings (Review Recommended)
1. [Warning] - [Location] - [Context]
### All Checks
- [ ] Type checking: [PASS/FAIL]
- [ ] Tests: [PASS/FAIL]
- [ ] Linting: [PASS/FAIL]
- [ ] Security: [PASS/FAIL]
- [ ] Build: [PASS/FAIL]
### Subagent Summary
- Initial findings: X issues
- After adversarial review: Y confirmed issues
- False positives filtered: Z
```
## Usage
This command ships with the project-starter plugin. Invoke with: `/project-starter:verify-changes`
© CloudAI-X · MIT · CloudAI-X/claude-workflow-v2