agentic-layer-assessment

Agentic Layer Assessment

Assess agentic layer maturity using the complete 12-grade classification system from TAC Lesson 14.

When to Use

• Evaluating current agentic layer maturity
• Identifying the next grade to achieve
• Tracking progress toward Codebase Singularity
• Onboarding new team members to agentic patterns
• Planning agentic infrastructure investments

Prerequisites

• Access to the codebase's .claude/ directory
• Understanding of @adw-framework.md classification system

The Classification System

Three classes with 12 total grades:

Class 1: Foundation (In-Loop Agentic Coding)

Grade	Component	Indicator
1	Memory Files	CLAUDE.md exists with guidance
2	Sub-Agents	Task agents used for parallelization
3	Skills/MCPs	Custom skills or MCP integrations
4	Closed-Loops	Self-validating prompts
5	Templates	Bug/feature/chore classification
6	Prompt Chains	Multi-step composite workflows
7	Agent Experts	Expertise files with self-improve

Class 2: External Integration (Out-Loop Agentic Coding)

Grade	Component	Indicator
1	Webhooks	External triggers (PITER framework)
2	ADWs	AI Developer Workflows running

Class 3: Production Orchestration (Orchestrated Agentic Coding)

Grade	Component	Indicator
1	Orchestrator	Meta-agent managing fleet
2	Orchestrator Workflows	Human-orchestrator interaction
3	ADWs + Orchestrator	Full autonomous execution

Assessment Process

Step 1: Scan Codebase

Check for indicators of each grade:

# Grade 1: Memory files
ls .claude/ CLAUDE.md

# Grade 2: Sub-agents
ls .claude/agents/

# Grade 3: Skills
ls .claude/skills/ || ls -d */skills/ 2>/dev/null

# Grade 4: Closed-loop patterns
grep -r "validation" .claude/commands/
grep -r "retry" .claude/commands/

# Grade 5: Templates
ls .claude/commands/ | grep -E "(chore|bug|feature)"

# Grade 6: Prompt chains
grep -r "Step 1" .claude/commands/
grep -r "Then execute" .claude/commands/

# Grade 7: Agent experts
ls .claude/commands/experts/ 2>/dev/null
find . -name "expertise.yaml"

# Grade 8 (Class 2 G1): Webhooks
find . -name "*webhook*" -o -name "*trigger*"

# Grade 9 (Class 2 G2): ADWs
ls adws/ 2>/dev/null

# Grade 10-12 (Class 3): Orchestrator
find . -name "*orchestrator*"

Step 2: Score Each Grade

For each grade, determine status:

Status	Meaning
✅ Complete	Fully implemented and used
🔶 Partial	Some elements present
❌ Missing	Not implemented

Step 3: Calculate Current Level

Your level = highest consecutive completed grade

Example:

• Grades 1-4: ✅
• Grade 5: 🔶
• Grades 6-7: ❌

Result: Class 1 Grade 4 (solid), targeting Grade 5

Step 4: Identify Next Step

Recommend specific actions for next grade:

Current	Next Step
Grade 1	Add Task agents for parallelization
Grade 2	Create custom skills or MCP
Grade 3	Add validation loops to prompts
Grade 4	Implement issue classification templates
Grade 5	Chain prompts into workflows
Grade 6	Build first agent expert
Grade 7	Set up external triggers
C2G1	Implement AI Developer Workflows
C2G2	Build orchestrator agent
C3G1	Add human-orchestrator workflows
C3G2	Connect orchestrator to ADWs

Output Format

## Agentic Layer Assessment Report

**Codebase:** [project name]
**Date:** [assessment date]
**Assessed by:** [model]

### Classification Summary

**Current Level:** Class [1/2/3] Grade [1-7/1-2/1-3]
**Maturity Score:** [X]/12 grades achieved

### Grade-by-Grade Assessment

| Grade | Component | Status | Evidence |
| --- | --- | --- | --- |
| C1G1 | Memory Files | ✅/🔶/❌ | [what was found] |
| C1G2 | Sub-Agents | ✅/🔶/❌ | [what was found] |
...

### Strengths

- [What's working well]

### Gaps

- [What's missing or weak]

### Recommended Next Steps

1. **Priority 1:** [Most impactful improvement]
2. **Priority 2:** [Second priority]
3. **Priority 3:** [Third priority]

### Path to Class 3

[Roadmap of remaining grades to achieve]

Assessment Checklist

• Scanned .claude/ directory structure
• Checked for memory files (CLAUDE.md)
• Searched for agent/skill definitions
• Analyzed prompt patterns (loops, chains)
• Looked for templates and classification
• Checked for expertise files
• Searched for external triggers
• Identified ADW presence
• Assessed orchestrator implementation
• Calculated maturity score
• Identified highest consecutive grade
• Recommended next steps

Key Insight

"Your agentic layer should be specialized to fit and wrap your codebase. Don't focus on reuse, focus on making these prompts great for that one codebase."

Each grade builds on the previous. Skip a grade and the foundation becomes unstable.

Anti-Patterns

Anti-Pattern	Problem	Solution
Skipping grades	Missing foundation	Build progressively
Over-engineering early	Complexity before value	Start with Grade 1-2
Generic layers	Don't fit codebase	Specialize for your project
Assessment without action	No improvement	Prioritize next step

Cross-References

• @adw-framework.md - Classification system details
• @agentic-layer-structure.md - Directory structure
• @zte-progression.md - Zero-touch engineering path
• @minimum-viable-agentic skill - Starting point

Version History

• v1.0.0 (2026-01-01): Initial release (Lesson 14)

---

Last Updated

Date: 2026-01-01 Model: claude-opus-4-5-20251101