Skip to content
Experiment 002 · Citadel Agentic Labs

Elektra Skills

Responsible AI Engineering & Execution Automation

A delivery-focused CLAUDE.md persona with governance hooks, RAI gates, and a 12-phase execution engine.Built across 66+ production sessions. 200+ PRs. Zero silent failures.

Works with 30+ AI coding agents

CCClaude Code
CuCursor
WsWindsurf
CpGitHub Copilot
ClCline
RcRoo Code
AiAider
CxCodex

WHAT IS ELEKTRA?

A delivery-focused AI persona that lives in your .claude/ directory

Elektra is a CLAUDE.md persona with shell hooks that automatically enforce governance. Install once — every session runs through structured phases with RAI gates.

After install — what you get
your-project/
├── CLAUDE.md              # Elektra persona + coding standards
├── .claude/
│   ├── settings.json      # Hook registry + enabled plugins
│   ├── hooks/
│   │   ├── session-init.sh        # Session lifecycle tracking
│   │   ├── cycle-guard.sh         # Plan→Build→Test→Review→Ship
│   │   ├── token-cap-guard.sh     # Context window warnings
│   │   ├── rai-check.sh           # RAI guardrail on edits
│   │   └── quality-gate.sh        # Async lint + format checks
│   └── skills/
│       ├── standard-orders/       # 12-phase execution engine
│       ├── responsible-ai/        # 7-pillar RAI governance
│       └── autoresearch/          # Autonomous iteration loop

How it works

1

Session starts

session-init.sh creates lifecycle state, tracks phase + tool calls

2

Before every edit

Cycle guard enforces Plan → Build → Test → Review → Ship. RAI check classifies sensitive files.

3

After every edit

Quality gate runs async lint checks. Token guard warns at 140K/170K/190K context.

4

At RAI gates (P2.5, P5.5)

Files classified into tiers. Critical paths get 7-pillar checklist. No silent bypasses.

USE CASES

/SO-godspeed

Ship a feature across multiple sessions

12-phase execution: Context → Review → Plan → Execute → QA → Code Review → Accept → Ship. Hooks enforce the sequence — you can’t skip QA.

/SO-godspeed-resume

Resume where you left off after a crash

Reads git status, branch, commits, plan checkboxes, and PR status to auto-detect your phase. If ambiguous, it asks.

/responsible-ai

Prevent hallucinated financial figures

7-pillar framework: LLMs NEVER compute financial figures (deterministic only). Citations mandatory. Confidence scores surfaced. Every fallback logged.

THE PROBLEM

Why governance?

AI wrote the code but nobody reviewed the AI

66% of developers skip review on AI-generated code. Silent hallucinations, missing edge cases, and unverified claims ship to production.

Solved by responsible-ai

Tests pass but the feature was never wired up

The #1 root cause of “done but not shipped.” Unit tests green, but the endpoint was never created. The button was never connected.

Solved by standard-orders

We shipped and data leaked across tenants

Silent failures are the most dangerous. No logs, no alerts, no confidence score. The LLM hallucinated a financial figure and nobody caught it.

Solved by rai-gates

We learned each of these the hard way.

— Citadel Agentic Labs

GET STARTED

One command. All skills.

Commercial Use

Free for individuals and non-commercial projects. Building a commercial product with these skills?

Get a commercial license