sap-crew-analysis

name	sap-crew-analysis
description	Multi-agent ABAP code analysis via local CrewAI pipeline. 7 specialized agents find errors, security flaws, performance bottlenecks, and Clean ABAP violations.
trigger	{"keywords":["analyze this ABAP","debug Z* code","find errors in program","review ZCL_* class","check code quality","audit ABAP security","optimize ABAP","improve SAP code","SAP crew analysis"],"intent":"Run multi-agent ABAP code analysis to find errors, security flaws, performance bottlenecks, and Clean ABAP violations.","patterns":["Z[A-Z]{3,}.*error","ZCL_.*review","ABAP.*(debug|analyze|audit)"]}

SAP Crew Analysis — 7-Agent ABAP Debug & Review

You give it ABAP source. It returns a structured report: errors found (severity-tagged), fix suggestions with code snippets, performance hotspots, security audit results, and a Clean ABAP compliance score. 7 agents, 13 tools, 4 phases. Runs locally — no data leaves your machine.

Prerequisites

# 1. Clone the crew agent repo (replace with actual URL)
git clone <your-sap-crew-agent-repo-url> ~/sap-crew-agent

# 2. Install Python dependencies
cd ~/sap-crew-agent/sap-crew-agent && pip install -r requirements.txt

# 3. Set LLM API key (any one works)
cp .env.example .env
# Edit .env → OPENAI_API_KEY=sk-... or DEEPSEEK_API_KEY=... or OPENROUTER_API_KEY=sk-or-...

# 4. Verify installation
python sap_crew/run.py --mode quick "Syntax check: is ABAP keyword CLASS valid?"

Requires: Python 3.10+, valid LLM API key, ADT MCP (arc-1/aibap) for source retrieval.

Quick Start

Analyze ABAP code (CLI)

# Full pipeline — security + quality + performance (3-5 min)
python sap_crew/run.py "Analyze ZCL_MATERIAL_HANDLER for bugs, security, and Clean ABAP compliance. Code: $(cat templates/ZROUTER_DISPATCH.abap | head -200)"

# Quick mode — fast bug hunt (1-2 min)
python sap_crew/run.py --mode quick "Review security of ZCL_MATERIAL_HANDLER: $(aibap get_class ZCL_MATERIAL_HANDLER)"

# Troubleshoot — diagnose a specific error (30s-1min)
python sap_crew/run.py --mode troubleshoot "CDS view ZI_PRODUCT has syntax error in join with MARA"

Via MCP Bridge (background, long-lived)

# Start MCP bridge once — stays alive, serves all requests
python sap_crew_mcp_bridge.py &

# Then call from any MCP client:
# sap_crew_delegate(task="Analyze ZCL_MATERIAL_HANDLER...", mode="full")

Analysis Modes

• full — Analysis → Design → Code → Review. 3-5 min. Use for complete security + quality audit.
• quick — Analysis → Code+Review. 1-2 min. Use for bug fixes and small improvements.
• troubleshoot — Diagnosis only. 30s-1min. Use for syntax errors and specific bugs.
• domain security — Security agent only. 2-3 min. Focused security audit.
• domain abap — ABAP-only agents. 2-3 min. ABAP-specific deep analysis.

The 7 Agents

• SAP Architect — Design review: architecture patterns, extensibility, design flaws.
• ABAP Developer — Code quality: Clean ABAP, BAPI patterns, modern syntax, inline declarations.
• HANA Specialist — DB performance: SELECT optimization, CDS perf, DB access patterns.
• CPI Expert — Integration flows: iFlow design, Groovy scripts, message mappings.
• CAP Developer — CAP services: CDS models, handlers, XSUAA auth, OData config.
• UI5/Fiori Dev — Frontend: UI5 patterns, Fiori guidelines, TypeScript compliance.
• Security Architect — Security audit: SQL injection, AUTHORITY-CHECK gaps, data privacy, SoD.

9 Analysis Dimensions

1. SEC (Security + ABAP) — SQL injection, dynamic code, OS calls, credential exposure.
2. AUTH (Security) — AUTHORITY-CHECK, S_DEVELOP, DCL roles, IAM gaps.
3. DATA (ABAP + HANA) — Error handling, ENQUEUE, COMMIT patterns, transaction safety.
4. PERF (HANA + ABAP) — SELECT in loop, SELECT *, missing indexes, full table scans.
5. STD (ABAP + Architect) — Deprecated statements, hardcoding, Clean ABAP violations.
6. INTERFACE (CPI + Architect) — RFC params, OData auth, timeout config, IDoc structure.
7. CHANGE (Architect) — Affected objects, SAP standard mods, shared INCLUDEs.
8. COMP (Security + Architect) — PII, dual-control, CDHDR/CDPOS logging, SoD.
9. FUNC (ABAP + Architect) — Requirements coverage, logic errors, edge cases.

ZROUTER Pipeline Integration

1. npm run abap:lint          ← abaplint static checks
2. sap-crew-analysis          ← 7-agent AI deep analysis (this skill)
3. npm run abap:review        ← combined lint + security gate
4. abap-code-review skill     ← structured 9-dimension report
5. sap-transport-gate skill   ← GO/NO-GO transport release gate

Example Output

## SAP Crew Analysis Report — ZCL_ROUTER_HANDLER_MM

### Errors Found (2 CRITICAL, 1 HIGH, 3 MEDIUM)

CRITICAL-1: BAPI_MATERIAL_SAVEDATA missing mandatory field
  Line: 421 | HEADDATA-MATERIAL_TYPE not set → BAPI returns E-type error
  Fix: ls_header-material_type = 'FERT'  " before BAPI call

CRITICAL-2: Missing BAPIRET2 table check
  Line: 426 | Only checks IMPORTING RETURN, misses TABLES RETURNMESSAGES
  Fix: READ TABLE lt_ret WITH KEY type = 'E'. IF sy-subrc = 0 OR ls_ret-type = 'E'.

HIGH-1: No ENQUEUE before BAPI
  Line: 420 | Concurrent users may hit lock errors
  Fix: Wrap in ENQUEUE_READ/ENQUEUE_WRITE block.

### Clean ABAP Compliance: 72/100
  ✅ Inline declarations (3/3)
  ✅ BAPI_TRANSACTION_COMMIT with WAIT (1/1)
  ❌ Long method: create_material is 67 lines (max 50)
  ❌ Missing method documentation
  ⚠️  SELECT * in get_material line 442

MCP Configuration

Add to .mcp.json:

{
  "sap-crew-local": {
    "type": "stdio",
    "command": "python",
    "args": ["/path/to/sap-crew-agent/sap_crew_mcp_bridge.py"],
    "description": "Local CrewAI — 7 agents, 13 tools, 4-phase pipeline. No cloud dependency."
  }
}

Pitfalls

• First run is slow — CrewAI initializes 7 LLM agents (~30s startup). Not an error.
• API costs scale with mode — quick ≈ 1K tokens; full ≈ 8K tokens. Choose troubleshoot for single-issue diagnosis.
• .env location matters — LLM key must be in sap-crew-agent/.env, not ambient shell env.
• Not a replacement for abaplint — Crew finds logic bugs (AI); abaplint finds syntax/pattern violations (deterministic). Use both.
• AI can hallucinate ABAP syntax — Always review suggestions before applying. The report is advisory, not authoritative.
• MCP bridge is long-lived — One instance serves all requests. Don't start multiple instances.
• Large source files — Truncate to ~200 lines for run.py CLI; full source via MCP bridge only.

Verification

# 1. Confirm crew agent is installed and responsive
python ~/sap-crew-agent/sap-crew-agent/sap_crew/run.py --mode quick "echo test"

# 2. Confirm MCP bridge starts without errors
python ~/sap-crew-agent/sap-crew-agent/sap_crew_mcp_bridge.py --health

# 3. Run a real analysis and check output structure
python sap_crew/run.py --mode troubleshoot "Check ZCL_TEST for syntax errors" | jq '.phases[] | {phase, status, findings}'

# 4. Verify report contains all 9 dimensions
python sap_crew/run.py --mode quick "Analyze: CLASS zcl_test DEFINITION. ENDCLASS." | grep -c 'SEC\|AUTH\|DATA\|PERF\|STD\|INTERFACE\|CHANGE\|COMP\|FUNC'
# Expected: ≥ 9 (one per dimension)

Performance & Resource Notes

• All processing is local — no source code leaves your machine.
• Full pipeline timeout: 10 min. Quick: 2 min. Troubleshoot: 1 min.
• ZROUTER dispatches to background thread — doesn't block main session.
• Same source + same request → crew reuses cached analysis when available.
• Monitor system resources with ST03N/SAT during large analyses (per SAP Performance Optimization Guide).