| name | review-team |
| description | Dispatch parallel specialized review agents simultaneously — code review, security audit, test coverage, and API contract review. Use when a feature/PR is ready for thorough review before merge. |
Review Team — Parallel PR Review
Overview
When a feature or PR is ready for review, a single sequential pass often misses issues that a specialist would catch. This skill dispatches four specialized reviewer agents in parallel, then aggregates their findings into a unified report with a single verdict.
Where this fits:
build feature → /review-team → address findings → merge
When to Use
Use when:
- A PR or feature branch is ready for final review before merge
- The change is non-trivial (new endpoints, auth changes, data migrations, new services)
- You want higher confidence before merging to canary/master
- A previous review found issues you want to verify are fully resolved
Do NOT use when:
- Quick typo fixes or single-line changes — standard review is sufficient
- Code is still WIP (use
/reviewing-code directly instead)
- You need a fast sanity check (use a single agent, not the full team)
- The change is purely to non-code artifacts (docs, configs only)
Step 1: Confirm Before Dispatch
Before dispatching any agents, summarize what will be reviewed:
Ready to dispatch the review team for: [PR/feature description]
Agents to be dispatched in parallel:
A. Code Reviewer — correctness, architecture, patterns (reviewing-code skill)
B. Security Reviewer — auth, injection, PII, guardrails G1-G11 (auditing-security skill)
C. Test Coverage — test existence, edge cases, missing scenarios
D. API Contract — naming, pagination, error responses (api-guidelines skill)
[SKIPPED if no API changes detected]
Estimated time: 2-4 minutes
Proceed? (yes/no)
Wait for explicit confirmation before dispatching.
Step 2: Detect API Changes
Before dispatching, check if the diff includes API changes:
- New or modified
@Controller, @RestController, @RequestMapping, @GetMapping, etc.
- New or modified endpoint paths
- Changes to request/response DTOs
If no API changes are detected, skip Agent D and note it in the report.
Step 3: Dispatch Agents in Parallel
Dispatch all agents simultaneously using the Agent tool. Each agent receives a focused prompt with the PR diff and relevant context.
Agent A — Code Reviewer
You are a code reviewer focused on correctness, architecture, and maintainability.
Invoke the reviewing-code skill and apply it to the following changes.
Focus on:
- Logic correctness and edge cases
- Architecture and design decisions (SOLID, separation of concerns)
- Error handling — are all exceptions caught and handled appropriately?
- Resource management — connections, streams, transactions properly closed?
- Idempotency — are mutations safe to retry?
- Code duplication and DRY violations
- Naming clarity and consistency
- Complexity: is there simpler way to achieve the same goal?
[PASTE DIFF / FILE PATHS HERE]
Return a structured list of findings. Each finding must include:
- severity: BLOCKING | IMPORTANT | SUGGESTION
- location: file + line number or method name
- issue: what is wrong
- recommendation: how to fix it
Agent B — Security Reviewer
You are a security auditor. Invoke the auditing-security skill and apply it to the
following changes. Check for myKaarma guardrails G1-G11.
Focus on:
- Authentication: are endpoints protected by mkid cookie auth where required?
- Authorization: is dealer context validated before operating on data?
- Injection vulnerabilities: SQL, NoSQL, command injection
- PII handling: are customer names, phones, emails logged or exposed incorrectly?
- Secrets: are any tokens, passwords, or API keys hardcoded or logged?
- Input validation: are all inputs sanitized before use?
- Insecure direct object references: are UUIDs validated against the caller's context?
- Rate limiting: are sensitive endpoints rate-limited?
[PASTE DIFF / FILE PATHS HERE]
Return a structured list of findings. Each finding must include:
- severity: BLOCKING | IMPORTANT | SUGGESTION
- guardrail: G1-G11 reference if applicable
- location: file + line number or method name
- issue: what is wrong
- recommendation: how to fix it
Agent C — Test Coverage Reviewer
You are a test coverage reviewer.
Examine the changed files and their corresponding test files. Check:
- Do tests exist for the changed functionality?
- Are happy-path cases covered?
- Are error and failure paths tested?
- Are edge cases covered (null inputs, empty lists, boundary values)?
- Are integration tests present for new API endpoints?
- Are external dependencies properly mocked in unit tests?
- Do tests assert behavior, not implementation details?
- Are there missing scenarios that could cause a production issue?
List specific untested scenarios by name, e.g.:
"Missing test: what happens when dealerUUID is not found?"
[PASTE DIFF / FILE PATHS HERE]
Return a structured list of findings. Each finding must include:
- severity: BLOCKING | IMPORTANT | SUGGESTION
- location: test file that should contain the test, or existing test file with gaps
- issue: what scenario is untested or poorly tested
- recommendation: what test to add
Agent D — API Contract Reviewer (if API changes detected)
You are an API contract reviewer. Invoke the api-guidelines skill and apply it to
the following API changes.
Focus on:
- URL patterns: plural nouns, no verbs, hyphens for multi-word terms
- HTTP methods: correct method for the operation
- Status codes: 4xx for client errors, 5xx for server errors, never 2xx with error body
- Error response format: structured JSON with code, message, details
- Versioning: all endpoints under /v1/ or higher
- Pagination: all list endpoints must have offset/limit
- Naming consistency: field names match org conventions
- Bulk operations: using :bulk-verb pattern with POST
- Auth parameters: in headers, not query params
- Documentation: Swagger/OpenAPI annotations present
[PASTE DIFF / FILE PATHS HERE]
Return a structured list of findings. Each finding must include:
- severity: BLOCKING | IMPORTANT | SUGGESTION
- location: file + line number or endpoint path
- issue: what violates the API guidelines
- recommendation: the correct pattern
Step 4: Aggregate and Report
After all agents return, perform aggregation:
Deduplication
- Collect all findings from all agents into one list
- Identify overlapping findings (e.g., Agent A and Agent B both flag the same missing auth check)
- Merge duplicates: keep the highest severity, reference both agents in the source
- Tag each finding with which agent(s) raised it
Severity Sort
Order findings: BLOCKING first, then IMPORTANT, then SUGGESTION.
Verdict
| Condition | Verdict |
|---|
| Any BLOCKING findings | REQUEST CHANGES |
| 3+ IMPORTANT findings with no BLOCKING | NEEDS DISCUSSION |
| Only IMPORTANT or SUGGESTION findings | REQUEST CHANGES or APPROVE (your judgment) |
| No findings above SUGGESTION level | APPROVE |
Output Format
Print the unified report in this structure:
## Review Team Report
PR / Feature: [name]
Date: [date]
Agents run: Code, Security, Test Coverage, API Contract (or note skipped)
### Verdict: [APPROVE | REQUEST CHANGES | NEEDS DISCUSSION]
---
### BLOCKING Issues ([count])
[B1] [severity] [Agent: A/B/C/D] — [file:line]
Issue: ...
Recommendation: ...
---
### IMPORTANT Issues ([count])
[I1] [severity] [Agent: A/B/C/D] — [file:line]
Issue: ...
Recommendation: ...
---
### SUGGESTIONS ([count])
[S1] ...
---
### What Looks Good
- [positive observations]
Then emit the structured JSON summary:
{
"verdict": "REQUEST_CHANGES | APPROVE | NEEDS_DISCUSSION",
"pr": "[pr name or branch]",
"reviewed_at": "[ISO date]",
"agents_run": ["code", "security", "test_coverage", "api_contract"],
"summary": {
"blocking": [count],
"important": [count],
"suggestions": [count]
},
"findings": [
{
"id": "B1",
"severity": "BLOCKING",
"agents": ["B"],
"location": "src/main/java/.../PaymentController.java:42",
"issue": "...",
"recommendation": "..."
}
]
}
This JSON can be pasted into a Jira comment or PR description automation.
Conflict Resolution
When agents return conflicting findings (e.g., Agent A says "simplify this method" but Agent C says "this method needs more tests before refactoring"):
- Present both perspectives
- Flag it as NEEDS DISCUSSION
- Do not silently pick one — let the engineer decide
Key Constraints
- This skill orchestrates other skills — it does not duplicate their logic
- Never auto-dispatch without user confirmation (HARD-GATE above)
- Never modify code — this skill is read-only
- If an agent times out or errors, note it in the report and flag those areas for manual review