From AI to Chitta
Tools for working with AI in a way that can be checked: Claude Code plugins, LLM security scanning, and frameworks for architecture, cost and practice.
From AI to Chitta — open
Tools for working with AI in a way that can be checked.
Everything here comes out of daily practice as an AI advisor: plugins that make Claude Code sessions disciplined and reproducible, security scanners that compute rather than guess, and frameworks for the parts of AI work that are usually asserted instead of measured. Built in the open, used every day.
New here? Start with voyage — brief, research, plan, execute, review — or llm-security, OWASP-aligned scanning for Claude Code projects.
Workflow and discipline for Claude Code
| Repo | What it does | Maturity |
|---|---|---|
| voyage | Contract-driven pipeline: brief, research, plan, execute, review | stable |
| config-audit | Multi-agent analysis and optimisation of Claude Code configuration | stable |
| graceful-handoff | One-command session handoff into the STATE.md continuity system | stable |
| repo-mailbox | Local mailbox for coordination between sessions in different repos | beta |
| repo-standard | Per-repo gate for this org's presentation standard | experimental |
| ktg-plugin-marketplace | The plugin catalog — install any of the above from here | stable |
Security
| Repo | What it does | Maturity |
|---|---|---|
| llm-security | Scanning, auditing and threat modeling for Claude Code projects | stable |
| llm-security-commons | Runtime-neutral core: detector data and normative contracts | beta |
| llm-ingestion-pipeline-security | Write-time defence for Python pipelines that persist LLM output | beta |
Architecture, cost and data
| Repo | What it does | Maturity |
|---|---|---|
| ms-ai-architect | Structured architecture guidance for the Microsoft AI stack | stable |
| portfolio-optimiser | Multi-agent cost optimisation on Microsoft Agent Framework, gated by a deterministic validator | experimental |
| portfolio-optimiser-commons | Framework-neutral shared core: normative specs and golden fixtures | experimental |
| llm-ingestion-okf | Open Knowledge Format ingestion library: connectors and bundle inbox | alpha |
Leadership and practice
| Repo | What it does | Maturity |
|---|---|---|
| okr | OKR guidance for Norwegian public sector, on 4-month tertial cycles | stable |
| ai-psychosis | Meta-awareness tools for healthy AI interaction patterns | stable |
Content and design
| Repo | What it does | Maturity |
|---|---|---|
| linkedin-studio | Content engine for Claude Code: posts, carousels, video, newsletters | beta |
| human-friendly-style | Shared plain-language output style for the marketplace | stable |
Maturity labels
Every repo carries a status-* topic, and the tables above read from it.
stable is in daily use and safe to depend on. beta works and is in use,
with an interface that may still move. alpha is early and will change.
experimental is research — read it, do not build on it yet.
Install
Most of the Claude Code work here ships as plugins. Add the marketplace once, then install what you want:
claude plugin marketplace add https://git.fromaitochitta.com/open/ktg-plugin-marketplace.git
claude plugin install voyage@ktg-plugin-marketplace
The first line only registers the catalog — it installs nothing on its own. The Python and framework repos are cloned and used directly; each has its own README.
License and security
MIT unless a repo says otherwise. Security contact: SECURITY.md.
More context and writing at fromaitochitta.com.