TECHNAI

TECHNAI is a shared library of agent skills, operating references, safety rules, and sync tooling for engineering teams that use Codex, Claude Code, Gemini CLI, GitHub Copilot CLI, Cursor, and related agent runtimes.

It gives each agent session a maintained set of skills, references, safety rules, and sync tooling. Features for continuous improvement from reviewed session evidence — audit, opportunity detection, and propagation — are described in the roadmap and are not yet implemented.

Name

TECHNAI is named after the ancient Greek technai, the plural of techne: arts, crafts, methods, practical knowledge, and disciplined know-how. In a Greco-Roman sense, it points to applied mastery: the structured arts and techniques that turn knowledge into repeatable work.

What TECHNAI Provides

• One owned source of truth for reusable SKILL.md packages.
• Multi-runtime exposure through native skill locations for Codex, Claude Code, Cursor, and shared ~/.agents/skills consumers.
• Governance rules for owned, adapted, reference-only, and excluded content.
• Agentic optimization roadmap (planned, not yet implemented) covering scope-aware context loading, session audit, opportunity detection, propagation, and cost reduction. See the roadmap.
• Reference library for fast-moving guidance across OpenAI/Codex, Claude Code, Gemini CLI, Cursor, security, prompt caching, and skills.
• Sync tooling with validation, backups, symlinks/junctions, copy-mode fallback, and dry-run support.

TECHNAI is not a vendor cache, credential store, model runtime, or scratchpad. Vendor/system/plugin-provided skills are tracked as references unless TECHNAI intentionally adopts and maintains them as owned content.

Repository Layout

TECHNAI/
├── skills/                  # Owned skill packages
├── scripts/                 # Validation and sync tooling
├── docs/                    # Product, governance, and operating library
├── docs/reference-library/  # Multi-system agentic reference library
├── registry/                # Source ownership and reference metadata
├── systemd/                 # Auto-update timer units
├── cron/                    # Cron fallback
├── tests/                   # Sync-tool regression tests
├── plugin.json              # Copilot CLI plugin manifest
└── README.md

Quick Start

Windows PowerShell:

git clone https://github.com/RBOKproject/TECHNAI.git C:\Dev\TECHNAI
Set-Location C:\Dev\TECHNAI
python scripts\agent-skills-sync --dry-run --link
python scripts\agent-skills-sync --link

Linux, WSL, or macOS:

git clone https://github.com/RBOKproject/TECHNAI.git ~/repos/TECHNAI
cd ~/repos/TECHNAI
python3 scripts/agent-skills-sync --dry-run --link
python3 scripts/agent-skills-sync --link

Expected healthy reruns report validate: PASS and ok for existing managed links.

Runtime Targets

The sync command validates every owned skill, then exposes the same source tree to:

• ~/.agents/skills
• ~/.codex/skills
• ~/.claude/skills
• ~/.cursor/skills

Gemini CLI and GitHub Copilot CLI can use ~/.agents/skills and may also be served through runtime-specific compatibility paths when needed.

Default exposure mode:

• Windows: directory junctions
• Linux, WSL, macOS: symlinks
• restricted hosts: explicit --link-mode copy

Existing unmanaged skill directories are moved to ~/.local/share/technai/backups/ before TECHNAI replaces them.

Documentation Library

Document	Purpose
Architecture	System boundaries, runtime model, and update flow.
Agent Fleet Launcher	Windows/WSL/WezTerm/tmux agent-fleet contract. Slot naming is project-neutral; the documented commands target ORDO profiles.
Compatibility	Supported runtimes, environments, and skill package rules.
Governance	Ownership, promotion, deprecation, and review requirements.
Onboarding	First install for Windows, WSL, Linux, macOS, and restricted hosts.
Operations	Sync, timers, diagnostics, rollback, and failure handling.
Usage Monitor	Operator CLI for provider quota, usage, reset, trend, and alert tracking.
Release Process	Checks, commit standards, publication, and release notes.
Security	Trust model, secret handling, supply chain, and safe execution.
Skill Standard	Authoring bar for maintainable, token-efficient skills.
Agentic Optimization Roadmap	Roadmap for planned scope-aware optimization features (not yet implemented).
Agentic Optimization Issue Pack	Epic and child issue list with DoD, dependencies, and references.
Reference Library	Current multi-system recommendations and freshness ledger.
Templates	Skill, review, and release-note templates.
External Assessment	Neutral evidence-and-maturity and valuation-input pack for independent external review (EN/FR).

Owned Skill Catalog

The current owned catalog covers:

• RBOK/ORDO delivery standards, issue-pack handoff, and dispatch exceptions
• RBOK frontend UX/UI implementation gates and Figma-first visual QA
• RBOK backend, testing, architecture, business, permissions, and deployment context
• local audio transcription hygiene for voice-note feedback
• Chrome DevTools MCP setup for browser audits and live Chrome workflows
• Windows + WSL workstation onboarding, repair, elevation, and hardening

Operating Principles

• Keep SKILL.md files concise and procedural.
• Move detailed, volatile, or variant-specific material into references.
• Classify content before sharing it: universal, organization, tool, project, user-private, or secret.
• Never store credentials, cookies, private keys, production secrets, or raw private session logs in TECHNAI.
• Use primary sources for current runtime guidance and refresh volatile references on cadence.
• Prefer reviewed propagation over ad hoc local prompt drift.

Validate Changes

Run these before publishing changes:

python -m unittest discover -s tests
python scripts/agent-skills-sync --dry-run --link
git diff --check

For runtime discovery changes, also run at least one relevant agent discovery check, such as gemini skills list all where available.

Auto Update

Shared Linux nodes can install the bundled systemd timer or cron entry to run:

agent-skills-sync --pull --link

The update path is fast-forward only. Do not force-reset a local TECHNAI checkout that may contain unreviewed skill or documentation work.