Qcsd Development Swarm

QCSD Development Swarm v1.0

Shift-left quality engineering swarm for in-sprint code quality assurance.

---

Overview

The Development Swarm takes refined stories (that passed Refinement) and validates code quality during sprint execution. Where the Ideation Swarm asks "Should we build this?" and the Refinement Swarm asks "How should we test this?", the Development Swarm asks "Is the code quality sufficient to ship?"

QCSD Phase Positioning

Parameters

• SOURCE_PATH: Source code directory to analyze (required, e.g., src/auth/)
• TEST_PATH: Test directory for coverage analysis (optional, e.g., tests/auth/)
• OUTPUT_FOLDER: Where to save reports (default: ${PROJECT_ROOT}/Agentic QCSD/development/)

---

ENFORCEMENT RULES - READ FIRST

PROHIBITED BEHAVIORS:

• Summarizing instead of spawning agents
• Skipping agents "for brevity"
• Proceeding before background tasks complete
• Providing your own analysis instead of spawning specialists
• Omitting report sections or using placeholder text

---

Step Execution Protocol

This skill uses a micro-file step architecture. Each step is a self-contained file loaded one at a time to avoid "lost in the middle" context degradation.

Execute steps sequentially by reading each step file with the Read tool.

Steps

1. Flag Detection -- steps/01-flag-detection.md -- Scan source code and tests, detect all 6 flags 2. Core Agents -- steps/02-core-agents.md -- Spawn qe-tdd-specialist, qe-code-complexity, qe-coverage-specialist in parallel 3. Batch 1 Results -- steps/03-batch1-results.md -- Wait for core agents, extract all metrics 4. Conditional Agents -- steps/04-conditional-agents.md -- Spawn flagged conditional agents in parallel 5. Decision Synthesis -- steps/05-decision-synthesis.md -- Apply SHIP/CONDITIONAL/HOLD logic 6. Report Generation -- steps/06-report-generation.md -- Generate executive summary and full report 7. Learning Persistence -- steps/07-learning-persistence.md -- Store findings to memory, save persistence record 8. Defect Predictor -- steps/08-defect-predictor.md -- Run qe-defect-predictor analysis on all code changes 9. Final Output -- steps/09-final-output.md -- Display completion summary with all scores

Execution Instructions

1. Use the Read tool to load the current step file (e.g., Read({ file_path: ".claude/skills/qcsd-development-swarm/steps/01-flag-detection.md" })) 2. Execute the step's instructions completely 3. Verify all success criteria are met before proceeding 4. Pass the step's output as context to the next step 5. If a step fails, halt and report the failure point -- do not skip ahead

Resume Support

To resume from a specific step: specify --from-step N and the orchestrator will skip to step N. Ensure you have the required prerequisite data from prior steps.

---

Agent Inventory

Total: 10 agents (3 core + 6 conditional + 1 analysis)

---

Quality Gate Thresholds

---

Report Filename Mapping

---

Execution Model Options

Workflow execution (ADR-102)

.claude/workflows/qcsd-development-review.js runs the review as a deterministic pipeline: one finder per quality dimension (TDD adherence, complexity, coverage gaps — args.dimensions selects a subset) → 3 blind adversarial refuters per finding (Loki-mode, ADR-074: refuters see only the bare claim + evidence, never the finder's confidence or each other; uncertainty defaults to refuted) → deterministic synthesis into finding-verdict@1 envelopes (ADR-103, schemas/finding-verdict.schema.json). A finding survives only if fewer than ⌈N/2⌉ refuters kill it. The final report contains ONLY confirmed findings; killed findings are retained under killed with their refutations for audit.

Args: sourcePath (required), testPath, dimensions (subset of tdd-adherence|complexity|coverage-gaps), maxFindings per dimension (default 5).

When the Workflow tool is unavailable, fall back to the Task-tool protocol below — the report format and gates are identical, minus the adversarial verification stage (note this in the report header as verification: none).

---

Key Principle

Code quality is measured by evidence, not intentions. This swarm provides in-sprint quality assessment to ensure code meets engineering standards before entering the CI/CD pipeline.

Step 1: Analyze Code Context (Flag Detection)

Prerequisites

• Development swarm skill invoked
• SOURCE_PATH provided
• OUTPUT_FOLDER determined

Instructions

Step 0: Retrieve Refinement Phase Signals

aqe memory search --pattern "qcsd-refinement-*" --namespace "qcsd-refinement" --limit 1 --json

Flag Detection (Check ALL SIX)

HAS_SECURITY_CODE = FALSE
  Set TRUE if source code includes: auth logic, crypto, input validation,
  token handling, password management, encryption, access control

HAS_PERFORMANCE_CODE = FALSE
  Set TRUE if source code includes: database queries, caching, async ops,
  batch processing, file I/O, network calls, concurrency

HAS_CRITICAL_CODE = FALSE
  Set TRUE if source code includes: payment processing, PII handling,
  financial calculations, healthcare data, safety-critical logic

HAS_MIDDLEWARE = FALSE
  Set TRUE if code includes: message broker clients, queue producers/consumers,
  ESB integration, event publishing, pub/sub patterns

HAS_SAP_INTEGRATION = FALSE
  Set TRUE if code includes: SAP SDK calls, IDoc processing, RFC calls,
  BAPI invocations, OData client code

HAS_AUTHORIZATION = FALSE
  Set TRUE if code includes: role checks, permission validation,
  SoD enforcement, RBAC logic, authorization middleware

Output flag detection results.

Success Criteria

• [ ] All SIX flags evaluated
• [ ] Expected agent count calculated

Navigation

• On success: proceed to Step 2 by reading steps/02-core-agents.md

Step 2: Spawn Core Agents (Parallel Batch 1)

Prerequisites

• Step 1 completed

Instructions

Spawn ALL THREE in ONE message

Agent 1: TDD Specialist

Analyze TDD adherence, test-code ratio, test-first evidence. Save to: ${OUTPUT_FOLDER}/02-tdd-analysis.md

Agent 2: Code Complexity

Analyze cyclomatic complexity, cognitive complexity, coupling. Save to: ${OUTPUT_FOLDER}/03-complexity-analysis.md

Agent 3: Coverage Specialist

Detect coverage gaps, untested paths, coverage by module. Save to: ${OUTPUT_FOLDER}/04-coverage-analysis.md

Success Criteria

• [ ] All THREE agents spawned

Navigation

• On success: proceed to Step 3 by reading steps/03-batch1-results.md

Step 3: Wait for Batch 1 & Extract Results

Prerequisites

• Step 2: All 3 core agents spawned

Instructions

WAIT for all three. Extract: TDD score, complexity metrics, coverage percentages, gaps identified.

Success Criteria

• [ ] All metrics extracted

Navigation

• On success: proceed to Step 4 by reading steps/04-conditional-agents.md

Step 4: Spawn Conditional Agents (Parallel Batch 2)

Prerequisites

• Step 3 completed, flags from Step 1

Instructions

If ALL flags FALSE, skip to Step 5.

Success Criteria

• [ ] All flagged agents completed

Navigation

• On success: proceed to Step 5 by reading steps/05-decision-synthesis.md

Step 5: Synthesize Results & Determine Recommendation

Prerequisites

• Core and conditional results available

Instructions

Apply SHIP/CONDITIONAL/HOLD logic EXACTLY

HOLD conditions (ANY triggers HOLD):

• Coverage < 50%
• Complexity score > critical threshold
• TDD adherence < 30%
• Security vulnerabilities (critical/high) in security code
• Mutation score < 40% on critical code

SHIP conditions (ALL required):

• Coverage >= 80%
• Complexity within acceptable range
• TDD adherence >= 70%
• No critical/high security findings
• All conditional checks passed

Default: CONDITIONAL

Success Criteria

• [ ] FINAL RECOMMENDATION: SHIP / CONDITIONAL / HOLD

Navigation

• On success: proceed to Step 6 by reading steps/06-report-generation.md

Step 6: Generate Development Report

Prerequisites

• Step 5 completed

Instructions

Generate the Development Report to ${OUTPUT_FOLDER}/01-executive-summary.md. Include: Executive Summary, TDD Analysis, Complexity Analysis, Coverage Analysis, conditional sections, Recommended Actions.

Success Criteria

• [ ] Complete report saved

Navigation

• On success: proceed to Step 7 by reading steps/07-learning-persistence.md

Step 7: Store Learnings & Persist State

Prerequisites

• Step 6 completed

Instructions

ALWAYS RUN.

Store development findings:

aqe memory store \
  --key "qcsd-development-${storyId}-${Date.now()}" \
  --namespace "qcsd-development" \
  --value '{...}' \
  --json

Save to ${OUTPUT_FOLDER}/11-learning-persistence.json.

Success Criteria

• [ ] Persistence completed

Navigation

• On success: proceed to Step 8 by reading steps/08-defect-predictor.md

Step 8: Apply Defect Predictor (Analysis)

Prerequisites

• Step 7 completed

Instructions

ALWAYS RUN qe-defect-predictor

Spawn qe-defect-predictor to analyze all code changes for defect likelihood:

1. Analyze code complexity hotspots 2. Predict defect density for changed modules 3. Identify highest-risk code paths 4. Recommend additional testing focus areas 5. Cross-reference with historical defect data

Save to: ${OUTPUT_FOLDER}/12-defect-prediction.md

Success Criteria

• [ ] Defect predictor completed
• [ ] Report saved

Navigation

• On success: proceed to Step 9 by reading steps/09-final-output.md

Step 9: Final Output & Completion Summary

Prerequisites

• All steps completed

Instructions

Display completion summary with: SHIP / CONDITIONAL / HOLD recommendation, TDD score, complexity metrics, coverage percentage, defect prediction, and all deliverables.

Success Criteria

• [ ] Summary displayed

Navigation

• Terminal step.