Search

PythonReactSEOPDF
GitHub Skill

zte-progression

Guide progression from In-Loop to Out-Loop to Zero-Touch Engineering. Use when assessing agentic maturity, planning ZTE progression, or identifying requirements for autonomous operation.

melodic-software☆ 21 Repository stars4 Repository forks

Installation

Follow the repository instructions for this skill and your agent. Review the files and dependencies before adding it to your workflow.

View source ↗Guides

Skill instructions

ZTE Progression Skill

Guide teams through the three levels of agentic coding: In-Loop, Out-Loop, and Zero-Touch Engineering.

When to Use

  • Assessing current agentic maturity level
  • Planning progression to higher automation
  • Identifying blockers to ZTE adoption
  • Building confidence for autonomous operation

Core Concepts

Three Levels
Level Presence KPI Description
In-Loop Constant Interactive prompting
Out-Loop 2 (prompt + review) AFK agents with review
Zero-Touch 1 (prompt only) Full automation
Progression Requirements

In-Loop -> Out-Loop:

  • Workflows succeed on first attempt
  • Tests catch issues reliably
  • Similar tasks repeat frequently

Out-Loop -> ZTE:

  • 90%+ success rate
  • Review catches nothing new
  • Tests provide safety net

Assessment Workflow

Step 1: Identify Current Level

Search for agentic workflow indicators:

Look for:
- Manual prompting patterns (In-Loop)
- Automated workflows with review (Out-Loop)
- Fully automated shipping (ZTE)
Step 2: Analyze Success Metrics

Review KPI data if available:

Check @agentic-kpis.md or equivalent:
- Attempt counts
- Success streaks
- Review catch rates
Step 3: Identify Blockers

Common blockers to progression:

Blocker Mitigation
Low test coverage Improve tests before progressing
Inconsistent success Analyze failures, improve prompts
Review catches issues Tests need to catch these first
No rollback capability Add before enabling ZTE
Step 4: Recommend Next Steps

Based on current level, recommend:

At In-Loop:

  • Start with Out-Loop for chores
  • Build workflow automation
  • Establish review process

At Out-Loop:

  • Track review catch rate
  • Build confidence with simple tasks
  • Consider ZTE for high-confidence areas

Near ZTE:

  • Enable for chores first
  • Expand progressively
  • Monitor continuously

Key Memory References

  • @zte-progression.md - Three levels definition
  • @zte-confidence-building.md - Building ZTE confidence
  • @agentic-kpis.md - KPI tracking
  • @composable-primitives.md - Workflow building blocks

Output Format

Provide assessment:

## ZTE Assessment

**Current Level:** [In-Loop | Out-Loop | ZTE]
**Problem Classes Assessed:** [chores, bugs, features]

### Indicators Found
- [Evidence of current level]

### Blockers to Progression
- [Identified blockers]

### Recommendations
1. [Specific next step]
2. [Specific next step]

### Target Timeline
- [Realistic progression timeline]

Anti-Patterns to Identify

  • Skipping levels (Out-Loop before tests exist)
  • All-or-nothing thinking (must ZTE everything)
  • Ignoring failure signals (pushing forward despite issues)
  • Review theater (reviewing without catching anything)

Version History

  • v1.0.0 (2025-12-26): Initial release

Last Updated

Date: 2025-12-26 Model: claude-opus-4-5-20251101

Examples

Read the usage examples in the source repository.

View source ↗

Source history

Indexed on
2026-01-20
Index updated
2026-02-04
Branch
main

Index and sync dates describe this directory. View the source repository for code history.

GitHub ↗

chat Comments (0)

chat_bubble_outline

No comments yet. Be the first to share your thoughts!