Context Engineering Integration Plan v2

Context Engineering Integration Plan v2

Executive Summary

This plan integrates the Agent-Skills-for-Context-Engineering repository (MIT license) into CODITECT-CORE, adding 5 new agents, 8 new skills, 3 new commands, 4 new scripts, 3 new workflows, and 1 new hook.

Source Repository: external/Agent-Skills-for-Context-Engineering Author: Muratcan Koylan License: MIT (fully compatible)

Impact Summary

Component Type	Current Count	New Additions	Final Count
Agents	130	+5	135
Skills	187	+8	195
Commands	144	+3	147
Scripts	236	+4	240
Workflows	1,149	+3	1,152
Hooks	41	+1	42
Total	1,892	+24	1,916

---

Phase 1: Foundation (Week 1)

1.1 Skill Integration (Priority: P0)

Adapt 8 skills from the external repository to CODITECT frontmatter standards.

Skill Name	Source	Priority	Complexity	Est. Hours
context-fundamentals	skills/context-fundamentals/	P0	Low	2
context-degradation	skills/context-degradation/	P0	Medium	3
context-compression	skills/context-compression/	P0	Medium	3
memory-systems	skills/memory-systems/	P1	High	4
context-optimization	skills/context-optimization/	P1	Medium	3
advanced-evaluation	skills/advanced-evaluation/	P1	High	4
tool-design	skills/tool-design/	P2	Low	2
project-development	skills/project-development/	P2	Medium	3

Adaptation Requirements:

1. Convert to CODITECT YAML frontmatter format (see templates)
2. Add component_type: skill and required identity fields
3. Add levels: section for progressive disclosure
4. Add when_to_use: and when_not_to_use: fields
5. Add composes_with: and requires: relationships
6. Preserve original Python scripts in scripts/ subdirectory
7. Move detailed references to references/ subdirectory

1.2 Script Adaptation (Priority: P0)

Create wrapper modules for utility functions to integrate with CODITECT patterns.

Script	Source	Integration Target	Est. Hours
context_health_monitor.py	degradation_detector.py	/cx command enhancement	3
compression_quality_test.py	compression_evaluator.py	Session compression	3
batch_pipeline.py	pipeline_template.py	Workflow automation	4
tool_docs_validator.py	description_generator.py	Component validation	2

---

Phase 2: Agent Creation (Week 2)

2.1 New Agent Definitions (Priority: P0)

Create 5 specialized agents leveraging external utility functions.

Agent Name	Domain	MoE Role	Base Utilities	Est. Hours
context-health-analyst	qa	analyst	degradation_detector.py	4
compression-evaluator	qa	judge	compression_evaluator.py	4
llm-judge	qa	judge	evaluation_example.py	4
multi-agent-coordinator	development	orchestrator	coordination.py	5
tool-designer	documentation	specialist	description_generator.py	3

Agent Template Enhancement:

• All agents will follow config/frontmatter-templates/agent-template.md
• Include Claude 4.5 optimization sections
• Reference skills via composes_with:
• Include invocation examples

2.2 Agent Integration Patterns

Each agent integrates with existing CODITECT patterns:

context-health-analyst
├── Invocation: /agent context-health-analyst "analyze session"
├── Composes with: context-degradation, context-fundamentals
├── Output: Health report with recommendations
└── Integration: Enhances /cx command

compression-evaluator
├── Invocation: /agent compression-evaluator "evaluate compression"
├── Composes with: context-compression, advanced-evaluation
├── Output: Probe-based quality scores
└── Integration: Session summarization workflow

llm-judge
├── Invocation: /agent llm-judge "evaluate response quality"
├── Composes with: advanced-evaluation
├── Output: Direct scoring, pairwise comparison, rubrics
└── Integration: Quality gates, PR reviews

multi-agent-coordinator
├── Invocation: /agent multi-agent-coordinator "coordinate workflow"
├── Composes with: memory-systems, context-optimization
├── Output: Orchestration plan with failure handling
└── Integration: Enhances orchestrator capabilities

tool-designer
├── Invocation: /agent tool-designer "improve command documentation"
├── Composes with: tool-design
├── Output: Validated tool descriptions
└── Integration: Component creation workflow

---

Phase 3: Command Integration (Week 3)

3.1 New Command Definitions (Priority: P1)

Command	Aliases	Description	Est. Hours
/context-health	/cx-health	Analyze session context health	3
/evaluate-response	/eval	LLM-as-judge evaluation	3
/batch-pipeline	/batch	Manage LLM batch processing	4

3.2 Command Specifications

/context-health

command_name: "/context-health"
aliases: ["/cx-health", "/health"]
usage: "/context-health [--verbose] [--recommendations]"
parameters:
  - name: verbose
    type: boolean
    default: false
    description: "Show detailed metrics"
  - name: recommendations
    type: boolean
    default: true
    description: "Include improvement recommendations"
requires_confirmation: false
modifies_files: false

Functionality:

• Token utilization analysis
• Attention degradation detection (lost-in-middle)
• Context poisoning indicators
• Health score (0-1) with status (healthy/warning/degraded/critical)
• Actionable recommendations

/evaluate-response

command_name: "/evaluate-response"
aliases: ["/eval", "/judge"]
usage: "/evaluate-response <mode> [options]"
parameters:
  - name: mode
    type: string
    required: true
    enum: [direct, pairwise, rubric]
    description: "Evaluation mode"
  - name: criteria
    type: string
    default: "accuracy,completeness,clarity"
    description: "Comma-separated evaluation criteria"
requires_confirmation: false
modifies_files: false

Functionality:

• Direct scoring with weighted criteria
• Pairwise comparison with position bias mitigation
• Rubric generation for domain-specific evaluation

/batch-pipeline

command_name: "/batch-pipeline"
aliases: ["/batch"]
usage: "/batch-pipeline <stage> --batch-id <id> [options]"
parameters:
  - name: stage
    type: string
    required: true
    enum: [acquire, prepare, process, parse, render, all, clean, estimate]
  - name: batch-id
    type: string
    required: true
  - name: workers
    type: number
    default: 5
requires_confirmation: true
modifies_files: true

---

Phase 4: Workflow & Hook Creation (Week 4)

4.1 New Workflows (Priority: P2)

Workflow	Category	Est. Hours
context-optimization-workflow.md	AI/ML	3
llm-evaluation-workflow.md	AI/ML	3
batch-processing-workflow.md	Automation	3

4.2 New Hook (Priority: P2)

Hook	Event	Description	Est. Hours
post-session-context-health	session_end	Auto-analyze context health	2

---

Phase 5: Template Enhancement (Week 4)

5.1 Proposed Template Improvements

Based on analysis of external repository patterns, propose backward-compatible enhancements:

Skill Template Enhancement

File: config/frontmatter-templates/skill-template.md

Additions (backward-compatible):

# NEW: Context Engineering Metadata (optional)
context_engineering:
  token_budget: ~3000
  attention_priority: beginning  # beginning | end | distributed
  compression_safe: true  # Safe to compress without loss

# NEW: Quality Metrics (optional)
quality_metrics:
  probe_tested: false  # Has passed probe-based evaluation
  coverage_score: 0  # % of use cases covered

Rationale: These additions support context-aware skill activation and quality validation without breaking existing skills.

Agent Template Enhancement

File: config/frontmatter-templates/agent-template.md

Additions (backward-compatible):

# NEW: Context Requirements (optional)
context_requirements:
  min_tokens: 0
  max_tokens: 8000
  requires_full_history: false
  attention_sensitive_regions: []

# NEW: Failure Handling (optional)
failure_handling:
  retry_strategy: exponential_backoff
  max_retries: 3
  circuit_breaker_threshold: 5
  fallback_agent: null

Rationale: Supports resilient multi-agent coordination with circuit breaker patterns.

5.2 Backward Compatibility Guarantee

All template changes are:

• ADDITIVE ONLY - No existing fields removed or renamed
• OPTIONAL - New fields have sensible defaults
• NON-BREAKING - Existing components work without modification
• VALIDATED - Schema validation passes for all existing components

---

Phase 6: Validation & Documentation (Week 5)

6.1 Validation Tasks

Task	Tool	Est. Hours
Validate all new YAML frontmatter	Schema validator	2
Update component-counts.json	update-component-counts.py	0.5
Update component-activation-status.json	update-component-activation.py	1
Run integration tests	test-suite.py	2
Update COMPONENT-REFERENCE.md	codi-documentation-writer	2

6.2 Documentation Updates

Document	Updates Required	Est. Hours
docs/reference/COMPONENT-REFERENCE.md	Add 24 new components	2
docs/guides/MEMORY-MANAGEMENT-GUIDE.md	Add context health section	1
docs/workflows/WORKFLOW-LIBRARY-INDEX.md	Add 3 new workflows	1
CLAUDE.md	Update component counts	0.5

---

Risk Assessment

Risk	Probability	Impact	Mitigation
Template incompatibility	Low	High	All changes additive/optional
Skill duplication	Medium	Low	Name disambiguation
Agent tool conflicts	Low	Medium	Tool validation before merge
Performance degradation	Low	Medium	Token budget monitoring

---

Success Criteria

Phase Completion Criteria

• Phase 1: 8 skills pass YAML validation
• Phase 2: 5 agents invocable via /agent
• Phase 3: 3 commands operational
• Phase 4: Workflows & hook functional
• Phase 5: Templates enhanced, backward compatible
• Phase 6: Documentation updated, tests passing

Quality Gates

• All components have valid YAML frontmatter
• All agents include Claude 4.5 optimization sections
• All skills include when_to_use and when_not_to_use
• Component counts updated in config/component-counts.json
• No breaking changes to existing components

---

Resource Allocation

Estimated Total Effort

Phase	Hours	Primary Agent
Phase 1: Skills	24	codi-documentation-writer
Phase 2: Agents	20	orchestrator
Phase 3: Commands	10	codi-documentation-writer
Phase 4: Workflows	11	orchestrator
Phase 5: Templates	4	project-organizer
Phase 6: Validation	9	qa-reviewer
Total	78

Agent Responsibilities

Agent	Primary Tasks
orchestrator	Phase coordination, agent creation
codi-documentation-writer	Skill/command documentation
project-organizer	File organization, template updates
qa-reviewer	Validation, quality gates
context-health-analyst (new)	Self-validation after creation

---

Execution Timeline

Week 1: Foundation
├── Day 1-2: Skill migration (4 skills)
├── Day 3-4: Skill migration (4 skills)
└── Day 5: Script adaptation

Week 2: Agents
├── Day 1-2: Agent definitions (3 agents)
├── Day 3-4: Agent definitions (2 agents)
└── Day 5: Agent testing

Week 3: Commands
├── Day 1-2: Command definitions
├── Day 3: Command testing
└── Day 4-5: Integration testing

Week 4: Polish
├── Day 1-2: Workflows & hook
├── Day 3: Template enhancements
└── Day 4-5: Validation

Week 5: Documentation
├── Day 1-3: Documentation updates
├── Day 4: Final validation
└── Day 5: Release

---

Appendix A: File Locations

New Skills

skills/
├── context-fundamentals/SKILL.md
├── context-degradation/SKILL.md
├── context-compression/SKILL.md
├── memory-systems/SKILL.md
├── context-optimization/SKILL.md
├── advanced-evaluation/SKILL.md
├── tool-design/SKILL.md
└── project-development/SKILL.md

New Agents

agents/
├── context-health-analyst.md
├── compression-evaluator.md
├── llm-judge.md
├── multi-agent-coordinator.md
└── tool-designer.md

New Commands

commands/
├── context-health.md
├── evaluate-response.md
└── batch-pipeline.md

New Scripts

scripts/context-engineering/
├── context_health_monitor.py
├── compression_quality_test.py
├── batch_pipeline.py
└── tool_docs_validator.py

New Workflows

docs/workflows/
├── context-optimization-workflow.md
├── llm-evaluation-workflow.md
└── batch-processing-workflow.md

New Hooks

hooks/
└── post-session-context-health.md

---

Appendix B: Integration Dependencies

---

Document Status: Active Last Updated: 2025-12-27 Author: Integration Planning Agent Approved By: Pending