Skillget

tenetx

agent
SUMMARY

The AI coding tool that adapts to you — personalized harness for Claude Code

README.md

Tenetx

The AI coding tool that adapts to you.

CI npm version License: MIT Tests: 1555

Quick Start · How It Works · Commands · Compound · 한국어 · 简体中文 · 日本語

What is Tenetx?

Other AI tools give everyone the same experience. Tenetx learns how you code and gets better over time.

$ tenetx forge                     # Profile your working style (once)
$ tenetx                           # Work normally — it learns silently

That's it. Tenetx extracts your coding patterns, validates them through real usage, and injects them when relevant. Bad patterns auto-retire. Good patterns get promoted.

Tenetx wraps Claude Code. It does not fork or modify it — it configures settings, hooks, and agents shaped by YOUR profile.

Quick Start

npm install -g tenetx
tenetx forge              # Scan your project + interview → done
tenetx                    # Work normally. Learning is automatic.

Prerequisites

  • Node.js >= 20
  • Claude Code installed and authenticated

    Tenetx wraps Claude Code and depends on its hook API. Future Claude Code updates may require tenetx updates.

When to Use Tenetx

Scenario Fit
Long-running project with repeating patterns Great
Personal workflow optimization Great
Lightweight harness (3 runtime deps) Great
One-off scripts or throwaway code Not ideal
Environment without Claude Code Not supported

How It Works

You code → Forge profiles you → Lab observes patterns → Compound extracts solutions
    ↓                                                          ↓
Config adapts (agents, skills, hooks, routing)     Solutions injected when relevant
    ↓                                                          ↓
Better AI assistance                               Good patterns promoted, bad ones retired

Four engines work together:

Engine What it does Command
Forge Profiles your working style across 5 dimensions tenetx forge
Lab Observes your behavior and auto-adjusts your profile tenetx lab
Compound Extracts coding patterns, validates through lifecycle tenetx compound
Pack Share and install community packs tenetx pack

Core Commands

# Essential
tenetx forge              # Profile your working style (once)
tenetx                    # Work normally (learning is automatic)

# When curious
tenetx me                 # See your profile + what it learned
tenetx compound list      # View learned patterns with status
tenetx pack search        # Browse community packs
tenetx pack install X     # Install a pack
tenetx pack publish X     # Share your pack

All 42 commands still work — run tenetx <command> --help for any command.

Forge — Your Profile

Forge scans your project and asks 10 questions to build a 5-dimension profile:

$ tenetx forge

Profile generated:
  품질 초점    [########--] 0.80    자율성 선호  [####------] 0.45
  위험 감수도  [######----] 0.62    추상화 수준  [#######---] 0.70
  커뮤니케이션 [#########-] 0.88

From your profile, Forge generates agent overlays, skill tuning, rules, hook parameters, model routing, and philosophy. Lab auto-adjusts this daily based on your actual behavior (EMA 0.25, max ±0.1/day).

Compound Engine — AI That Learns From You

Compound automatically extracts coding patterns, validates them through evidence, and shares them.

Session work → Auto-extraction (git diff) → Quality gates → Solution stored
    ↓
Next session → Tag matching → Solution injected → Code Reflection
    ↓
Identifier in your code → reflected++  |  Build failure → negative++
    ↓
Lifecycle check → Promote / Demote / Circuit-break / Retire

Evidence-Based Lifecycle

Solutions earn trust through real usage:

Status Confidence How to reach
experiment 0.3 Auto-extracted from git diff
candidate 0.6 reflected >= 2, sessions >= 2, negative == 0
verified 0.8 reflected >= 4, sessions >= 3, OR --verify
mature 0.85 reflected >= 8, sessions >= 5, sustained 30 days

What Makes It Different

  • Code Reflection — detects when Claude actually uses your pattern via identifier matching
  • Negative signals — build/test failures automatically demote bad patterns
  • Circuit breaker — experiment patterns with 2+ failures are auto-retired
  • Contradiction detection — flags patterns with overlapping tags but disjoint identifiers
  • Prompt injection defense — 13 patterns + Unicode NFKC + XML escaping
tenetx compound --solution "ErrorBoundary" "centralized error handling"
tenetx compound list                          # All patterns with status
tenetx compound inspect <name>                # Detailed evidence
tenetx compound --verify <name>               # Manual promote
tenetx compound --lifecycle                   # Run promotion check
tenetx compound rollback --since 2026-03-20   # Undo extractions

Pack Marketplace — Share Patterns

Publish and install community packs. No server — powered by GitHub.

tenetx pack search "react"           # Search community registry
tenetx pack install react-patterns   # Install a pack
tenetx pack publish my-pack          # Share your pack to registry

Packs bundle skills, agents, rules, and workflows. Sorted by download count.

Registry: wooo-jin/tenetx-registry

Advanced Features

Execution Modes (9 modes)
Flag Mode What it does
-a autopilot 5-stage autonomous pipeline
-r ralph PRD-based completion guarantee
-t team Multi-agent parallel pipeline
-u ultrawork Maximum parallelism burst
-p pipeline Sequential processing
ccg 3-model cross-validation
ralplan Consensus design (Planner→Architect→Critic)
deep-interview Socratic requirements clarification
tdd Test-driven development

Magic keywords work too — type autopilot, ralph, ultrawork, tdd anywhere in your prompt.

19 Dimension-Tuned Agents
Lane Agents Purpose
BUILD explore, analyst, planner, architect, debugger, executor, verifier, code-simplifier, refactoring-expert Exploration → Implementation → Verification
REVIEW code-reviewer, security-reviewer, critic Quality assurance
DOMAIN designer, test-engineer, writer, qa-tester, performance-reviewer, scientist, git-master Specialized expertise

A developer with Quality=0.80 gets strict code review. Quality=0.40 gets correctness-only review.

Model Routing (16-Signal Scoring) 
Haiku   →  explore, file-search, simple-qa
Sonnet  →  code-review, analysis, design
Opus    →  implement, architect, debug-complex

Your Forge profile adds overrides. Lab auto-adjusts routing based on your patterns.

Remix — Cherry-Pick From Others 
tenetx remix browse
tenetx remix pick strict-reviewer --from senior-dev-harness

Cherry-pick agents, skills, rules from published harnesses with conflict detection.

Multi-Model Synthesis 
tenetx ask "saga or choreography?" --all

Ask all configured providers, get confidence-scored synthesis.

txd — Skip Permissions Shortcut 
txd                   # Equivalent to: tenetx --dangerously-skip-permissions

Warning: txd disables ALL Claude Code permission checks. Tools execute without confirmation. Only use in trusted, isolated environments (e.g., disposable containers, personal dev machines with no sensitive data). Never use in production or shared environments.

Architecture

Tenetx Architecture

Layer Purpose Components
Profile Who you are 5 continuous dimensions
Adapt Learn and adjust Forge + Lab + Compound
Execute Do the work 9 modes, 21 skills, 19 agents, 17 hooks
Share Community Pack marketplace + Remix

Statistics

Metric Count
Tests 1855 across 107 files
Agents 19 (dimension-tuned)
Skills 21 (6 dimension-aware)
Compound quality gates 4
Lab event types 15
Execution modes 9
CLI commands 5 core + 37 advanced

Acknowledgements

Tenetx draws significant inspiration from oh-my-claudecode by Yeachan Heo. The multi-agent orchestration patterns, magic keyword system, and the vision of enhancing Claude Code through a harness layer were deeply influenced by OMC's pioneering work.

Where Tenetx diverges: OMC gives you powerful, general-purpose tools. Tenetx makes those tools personal. Forge profiles how you work. Lab evolves it. Compound Engine tracks which patterns actually help — promoting good ones and auto-retiring bad ones through evidence-based lifecycle.

Contact

License

MIT

Yorumlar (0)

Sonuc bulunamadi