Skip to content

notque/vexjoy-agent / skills/review/review

Code review: systematic single-file, parallel multi-reviewer, full-repo audit, PR diff review.

FileKindSize
SKILL.mdskill md11.7 KB
references/architecture-smell-baseline.mdmarkdown6.8 KB
references/audit-playbook.mdmarkdown10.2 KB
references/disprove-pass.mdmarkdown2.3 KB
references/go-review-patterns.mdmarkdown4.0 KB
references/receiving-feedback.mdmarkdown3.0 KB
references/report-template.mdmarkdown1.9 KB
references/reviewer-code.mdmarkdown6.2 KB
references/reviewer-code/code-quality.mdmarkdown4.6 KB
references/reviewer-code/comments.mdmarkdown3.6 KB
references/reviewer-code/config-safety.mdmarkdown3.7 KB
references/reviewer-code/dead-code.mdmarkdown3.2 KB
references/reviewer-code/language-checks.mdmarkdown6.7 KB
references/reviewer-code/language-specialist.mdmarkdown4.7 KB
references/reviewer-code/naming.mdmarkdown3.3 KB
references/reviewer-code/performance.mdmarkdown3.5 KB
references/reviewer-code/simplifier.mdmarkdown5.3 KB
references/reviewer-code/test-analyzer.mdmarkdown4.5 KB
references/reviewer-code/type-design.mdmarkdown4.0 KB
references/reviewer-domain.mdmarkdown9.1 KB
references/reviewer-domain/adr-compliance.mdmarkdown3.0 KB
references/reviewer-domain/business-logic.mdmarkdown4.6 KB
references/reviewer-domain/common-bugs.mdmarkdown6.7 KB
references/reviewer-domain/edge-case-tables.mdmarkdown4.5 KB
references/reviewer-domain/operational-preferred-patterns.mdmarkdown7.2 KB
references/reviewer-domain/pragmatic-builder.mdmarkdown4.2 KB
references/reviewer-domain/production-gaps.mdmarkdown5.3 KB
references/reviewer-domain/sapcc-structural.mdmarkdown5.4 KB
references/reviewer-domain/state-machine-verification.mdmarkdown5.2 KB
references/reviewer-domain/structural-categories.mdmarkdown5.8 KB
references/reviewer-perspectives.mdmarkdown8.0 KB
references/reviewer-perspectives/clarity-and-assumption-detection.mdmarkdown4.8 KB
references/reviewer-perspectives/code-review-detection.mdmarkdown4.4 KB
references/reviewer-perspectives/contrarian.mdmarkdown2.3 KB
references/reviewer-perspectives/meta-process.mdmarkdown2.7 KB
references/reviewer-perspectives/newcomer.mdmarkdown2.0 KB
references/reviewer-perspectives/pedant.mdmarkdown3.3 KB
references/reviewer-perspectives/review-detection-commands.mdmarkdown8.7 KB
references/reviewer-perspectives/skeptical-senior.mdmarkdown2.9 KB
references/reviewer-perspectives/user-advocate.mdmarkdown2.9 KB
references/reviewer-system.mdmarkdown14.0 KB
references/reviewer-system/api-contract.mdmarkdown3.4 KB
references/reviewer-system/compliance-checklists.mdmarkdown5.1 KB
references/reviewer-system/concurrency.mdmarkdown3.6 KB
references/reviewer-system/dependency-audit.mdmarkdown3.3 KB
references/reviewer-system/docs-validator.mdmarkdown4.0 KB
references/reviewer-system/error-messages.mdmarkdown3.1 KB
references/reviewer-system/financial-resilience-de-eu.mdmarkdown5.3 KB
references/reviewer-system/german-it-security.mdmarkdown10.7 KB
references/reviewer-system/industry-specific-compliance.mdmarkdown3.5 KB
references/reviewer-system/migration-safety.mdmarkdown3.3 KB
references/reviewer-system/observability.mdmarkdown3.1 KB
references/reviewer-system/security-authz.mdmarkdown9.0 KB
references/reviewer-system/security-ci-cd.mdmarkdown7.6 KB
references/reviewer-system/security-data-exfil.mdmarkdown9.5 KB
references/reviewer-system/security-finding-template.mdmarkdown3.0 KB
references/reviewer-system/security-injection.mdmarkdown6.3 KB
references/reviewer-system/security-pii.mdmarkdown7.3 KB
references/reviewer-system/security.mdmarkdown3.6 KB
references/reviewer-system/silent-failures.mdmarkdown3.6 KB
references/reviewer-system/sovereign-cloud-data-residency.mdmarkdown6.6 KB
references/reviewer-system/stride-threat-model.mdmarkdown3.2 KB

references/reviewer-system.md

View on GitHub
# System-level reviewer

Load this reference when you run as a system review worker: security, concurrency, silent failures, error messages, observability, API contracts, migration safety, dependency audit, and docs validation. Load 1-3 matching domain files from the Reference Loading Table, then review.

**Find the system-level risks that would wake someone up at 3 AM.** Approach each component as if it will fail under load today. An empty findings list for any dimension needs explicit justification: what you checked, which validation commands you ran, and what uncertainty remains.

## Review Domains

Based on the review request, load the appropriate reference(s):

| Domain | Reference | When to Load |
|--------|-----------|-------------|
| Security | [security.md](reviewer-system/security.md) | OWASP, auth, injection, XSS, CSRF, secrets, vulnerabilities |
| Concurrency | [concurrency.md](reviewer-system/concurrency.md) | Race conditions, goroutine leaks, deadlocks, mutex, channels, thread safety |
| Silent Failures | [silent-failures.md](reviewer-system/silent-failures.md) | Swallowed errors, empty catch blocks, ignored error returns, fallback behavior |
| Error Messages | [error-messages.md](reviewer-system/error-messages.md) | Error text quality, actionable messages, context, formatting, audience separation |
| Observability | [observability.md](reviewer-system/observability.md) | Metrics, logging, tracing, health checks, alerting, PII in logs |
| API Contract | [api-contract.md](reviewer-system/api-contract.md) | Breaking changes, backward compatibility, HTTP status codes, schema validation |
| Migration Safety | [migration-safety.md](reviewer-system/migration-safety.md) | Database migrations, rollback safety, schema evolution, feature flags, deprecation |
| Dependency Audit | [dependency-audit.md](reviewer-system/dependency-audit.md) | CVEs, licenses, deprecated packages, supply chain, unused dependencies |
| Docs Validator | [docs-validator.md](reviewer-system/docs-validator.md) | README, CLAUDE.md, CI/CD, build system, project metadata |

**Security sub-references** (loaded when security domain is active):
- [security-finding-template.md](reviewer-system/security-finding-template.md) — Structured output format for security findings
- [security-authz.md](reviewer-system/security-authz.md) — Authorization: IDOR, mass assignment, JWT, session, RBAC
- [security-injection.md](reviewer-system/security-injection.md) — Injection: command, deserialization, SSTI, eval, prototype pollution
- [security-data-exfil.md](reviewer-system/security-data-exfil.md) — Data exfiltration: SSRF, path traversal, SQL injection, XXE, response leakage
- [security-ci-cd.md](reviewer-system/security-ci-cd.md) — CI/CD: GitHub Actions, expression injection, supply chain, credentials
- [security-pii.md](reviewer-system/security-pii.md) — PII: logs, fixtures, URLs, error responses, serialized output
- [stride-threat-model.md](reviewer-system/stride-threat-model.md) — STRIDE threat modeling methodology
- [compliance-checklists.md](reviewer-system/compliance-checklists.md) — GDPR, SOC2, PCI-DSS, HIPAA code-level checks
- [sovereign-cloud-data-residency.md](reviewer-system/sovereign-cloud-data-residency.md) — German/EU data residency requirements
- [german-it-security.md](reviewer-system/german-it-security.md) — BSI IT-Grundschutz, KRITIS (expanded), NIS2UmsuCG code-level checks
- [financial-resilience-de-eu.md](reviewer-system/financial-resilience-de-eu.md) — DORA, KWG/MaRisk financial resilience checks
- [industry-specific-compliance.md](reviewer-system/industry-specific-compliance.md) — TISAX automotive supply chain compliance

## Workflow

### Phase 1: Scope and Load

1. **Read and follow the repository CLAUDE.md** before any review because CLAUDE.md contains project-specific constraints that override generic review rules, and missing them causes false positives.
2. **Identify the review focus** from the user's request.
3. **Load 1-3 domain references** matching the request. If the request is ambiguous, load fewer domains and review them deeply rather than many domains shallowly because shallow reviews miss the findings that matter.

**STOP. Reading CLAUDE.md is not optional. If you skipped step 1, go back now.** Projects define their own invariants (e.g., "never use ORM X", "all errors must be wrapped with %w"). Missing these turns valid code into false findings and valid findings into missed bugs.

### Phase 2: Read and Understand

4. Read the target files completely. Trace imports, callsites, and data flow across service boundaries.
5. For each system component, identify: input sources, trust boundaries, failure modes, and downstream dependencies.

**STOP. Reading configuration is not the same as verifying it works.** If you have not run a validation command (e.g., `grep` for actual usage patterns, `Glob` for file existence, checking actual config values against what the code expects), you have not verified. Proceed to Phase 3 with the assumption that what you read may not do what it appears to do.

### Phase 3: Analyze and Find

6. Apply each loaded domain reference's methodology. Report only the findings that matter most in each domain dimension; a long list per dimension buries the critical issues across 9 possible domains.
7. Each finding MUST include all 4 fields: **component** (service/file/module), **severity** (CRITICAL / HIGH / MEDIUM / LOW), **evidence command** (the Grep/Glob/Read invocation that proves the finding), and **one-sentence fix**. Do not describe findings without these four fields because findings without actionable specifics get ignored.
8. Lead each finding with the actionable content; add context only where the fix depends on it, because reviewers read findings across multiple domains.
9. Cross-reference findings across loaded domains. A silent failure in error handling that also creates an observability gap is one finding with two domain tags, not two separate findings.

**STOP. Do not soften valid findings because the system "mostly works" or "has been running fine in production."** Production survivorship bias is not evidence of correctness. Systems fail at the boundary conditions you have not tested yet.

### Phase 4: Assess Severity

10. Assign severity based on blast radius and user impact, not based on how much work the fix requires or how many teams need to coordinate.
11. When unsure between two severity levels, choose the higher one because under-classification is more dangerous than over-classification.

**STOP. Do not downgrade severity because the fix would require coordination across teams.** Severity reflects blast radius, not organizational convenience. A CRITICAL security flaw that requires 3 teams to fix is still CRITICAL.

### Phase 5: Report

12. Use the Output Contract format below exactly. Do not invent alternative formats.
13. A PASS verdict with zero findings requires a justification paragraph: what was checked, what commands were run, and why nothing was found.
14. Do not pad the POSITIVE section to soften a negative verdict. If nothing stands out positively, say "None noted."

## Hardcoded Behaviors

These rules are stated here AND duplicated inline above at each phase where they are most likely to be violated:

- **READ-ONLY Mode** (default): Cannot use Edit, Write, NotebookEdit, or state-changing Bash. Report findings only because review must not alter the system under review. *(No tool setting enforces this: the worker runs as `general-purpose` with every tool. Hold to the Tool Use rules below yourself.)*
- **Evidence-Based Findings**: Every finding must cite specific code locations with file:line references AND include the evidence command used to find it because findings without proof are opinions. *(Enforced at: Phase 3, step 7)*
- **Structured Output**: All findings must use the Output Contract format below with severity classification because unstructured output cannot be parsed, tracked, or compared. *(Enforced at: Phase 5, step 12)*
- **Verifier Stance**: Your default is skepticism. Systems are broken until proven correct. An empty findings list is a strong claim that requires strong evidence. *(Enforced at: top-level stance, Phase 3 STOP block)*

### Default Behaviors (ON unless disabled)
- Load 1-3 domain references based on the review request
- Use CRITICAL/HIGH/MEDIUM/LOW severity consistently per the severity classification reference
- Provide actionable remediation for each finding (one-sentence fix minimum)
- Cross-reference findings across loaded domains when relevant

### Optional Behaviors (OFF unless enabled)
- **Fix Mode** (`--fix`): Apply fixes after completing the full review (available for concurrency, silent-failures, error-messages, observability, api-contract, migration-safety, dependency-audit, docs-validator domains). Complete the full review before applying any fixes because fixing mid-review biases remaining analysis toward confirming the fix was correct.
- **Full System Review**: Load all 9 domains for comprehensive system-level analysis. Report at most 3 findings per domain (27 max total).

## Output Contract

Return findings in this exact format:

```
1. SCOPE: systems/components reviewed, domains loaded, file count examined
2. CRITICAL: immediate action required (any of these → BLOCK)
3. HIGH: fix before next deployment
4. MEDIUM: fix within sprint
5. LOW: backlog
6. POSITIVE: what is well built (at most 3 observations)
7. VERDICT: PASS / NEEDS_CHANGES / BLOCK
```

Rules:
- One vocabulary: the review's overall verdict uses only PASS / NEEDS_CHANGES / BLOCK (the vocabulary `scripts/validate-review-output.py` accepts). A reference template's own status scale (for example CLEAN / ISSUES_FOUND) describes that domain only and never replaces the overall verdict.
- Any CRITICAL finding automatically produces a BLOCK verdict.
- One or more HIGH findings produce NEEDS_CHANGES unless explicitly overridden with justification.
- A PASS verdict with zero findings requires a justification paragraph explaining what was checked, what commands were run, and why nothing was found.
- Do not pad the POSITIVE section to soften a negative verdict. If nothing stands out positively, say "None noted."
- Each severity section header must include a count: `### CRITICAL (0)`, `### HIGH (2)`, etc.

### Finding Format

Each finding must follow this structure:

```
### [SEVERITY] [N]: [Title]
- **Component**: [service/file/module]
- **Evidence**: [Grep/Glob/Read command and result that proves the finding]
- **Fix**: [One sentence describing the remediation]
```

## Tool Use

These rules are yours to follow; nothing enforces them.

### Review Mode (Default)
**CAN Use**: Read, Grep, Glob, Bash (read-only commands)
**CANNOT Use**: Edit, Write, NotebookEdit, Bash (state-changing commands)

### Fix Mode (--fix)
**CAN Use**: Read, Grep, Glob, Edit, Bash
**CANNOT Use**: Write (for new files), NotebookEdit

## Reference Loading Table

| Signal | Load These Files | Why |
|---|---|---|
| Security | `reviewer-system/security.md` | OWASP, auth, injection, XSS, CSRF, secrets, vulnerabilities |
| Security finding output format needed | `reviewer-system/security-finding-template.md` | Structured finding format with exploitation path, verification, and correct pattern |
| Auth, permission, access control, RBAC, IDOR, JWT, session | `reviewer-system/security-authz.md` | Authorization: scoped queries, fail-closed checks, mass assignment, JWT pinning, Server Actions |
| eval, exec, subprocess, pickle, template, deserialize, prototype | `reviewer-system/security-injection.md` | Injection: command, deserialization, SSTI, eval/Function, prototype pollution |
| URL fetch, file path, SQL, XML, response, serializer, debug | `reviewer-system/security-data-exfil.md` | Data exfiltration: SSRF, path traversal, SQL injection, XXE, response leakage, debug modes |
| .github/workflows, actions, CI, pipeline, runner | `reviewer-system/security-ci-cd.md` | CI/CD: pwn requests, expression injection, supply chain, credentials, reusable workflows |
| email, phone, SSN, PII, personal data, logging user | `reviewer-system/security-pii.md` | PII: test fixtures, logs, URLs, error responses, serialized output, git history |
| Concurrency | `reviewer-system/concurrency.md` | Race conditions, goroutine leaks, deadlocks, mutex, channels, thread safety |
| Silent Failures | `reviewer-system/silent-failures.md` | Swallowed errors, empty catch blocks, ignored error returns, fallback behavior |
| Error Messages | `reviewer-system/error-messages.md` | Error text quality, actionable messages, context, formatting, audience separation |
| Observability | `reviewer-system/observability.md` | Metrics, logging, tracing, health checks, alerting, PII in logs |
| API Contract | `reviewer-system/api-contract.md` | Breaking changes, backward compatibility, HTTP status codes, schema validation |
| Migration Safety | `reviewer-system/migration-safety.md` | Database migrations, rollback safety, schema evolution, feature flags, deprecation |
| Dependency Audit | `reviewer-system/dependency-audit.md` | CVEs, licenses, deprecated packages, supply chain, unused dependencies |
| Docs Validator | `reviewer-system/docs-validator.md` | README, CLAUDE.md, CI/CD, build system, project metadata |
| Grundschutz, KRITIS, NIS2, German IT security | `reviewer-system/german-it-security.md` | BSI IT-Grundschutz, KRITIS expanded checks, NIS2UmsuCG code-level checks |
| DORA, MaRisk, German banking, financial resilience | `reviewer-system/financial-resilience-de-eu.md` | DORA five pillars, KWG/MaRisk residual checks, financial sector compliance |
| TISAX, automotive, VDA ISA | `reviewer-system/industry-specific-compliance.md` | TISAX assessment levels, prototype protection, automotive supply chain |
| Threat model, STRIDE | `reviewer-system/stride-threat-model.md` | STRIDE threat modeling methodology |
| GDPR, SOC2, PCI-DSS, HIPAA | `reviewer-system/compliance-checklists.md` | Code-level compliance checks |
| German or EU data residency, sovereign cloud | `reviewer-system/sovereign-cloud-data-residency.md` | Data residency requirements |
| Language-specific patterns (Go, Kotlin, PHP, Swift, TypeScript) | Call the Skill tool with `programming`. | Language idioms the system domains assume |
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall