# mellicci.dev

> The portfolio, field notes and learning space of Martin Melišek: open-source tools, proprietary products, notes on agent tooling and token economics, and a free course on coding agent fundamentals — each linked to the code, package or writing behind it wherever one exists.

This file is generated at build time from the same content collections the site's pages render from, so it lists exactly the published projects and posts — nothing is maintained by hand. Every link is the entry's own page on this site.

## Projects

- [Agent Tokenwatch](https://mellicci.dev/work/agent-tokenwatch): A dependency-free Node.js CLI that captures local, metadata-only token and cost telemetry for Claude Code, Codex CLI, and GitHub Copilot CLI. Installs lifecycle hooks where available, feeds supported status lines, and ships a set of tw-* skills.
- [Atlas Data Fabric](https://mellicci.dev/work/atlas-data-fabric): The governed workspace for core-system data migrations. Agents and humans share the same features but not the same authority: agents propose, people approve, and every decision is audited. Approved mappings become the specs ETL is built from. Proprietary, available through Itera.
- [Code Compass](https://mellicci.dev/work/code-compass): Builds a local digital twin of a whole codebase, modern or mainframe, that agents query through local MCP tools. Precise callers, definitions and blast radius instead of noisy grep. Proprietary, available through Itera.
- [Coding Agent Fundamentals](https://mellicci.dev/work/coding-agent-fundamentals): A free, vendor-neutral course on how coding-agent harnesses work and how to build on them safely.
- [Digital Factory](https://mellicci.dev/work/digital-factory): Design, run, monitor, orchestrate and share multi-agent workflows. Manual prompting becomes versioned, auditable pipelines that scale a senior's delivery method across teams, interactively or in CI. Proprietary, available through Itera.
- [Meridian Mainframe Insight](https://mellicci.dev/work/meridian-mainframe-insight): Turns undocumented mainframe services into living documentation, delivered either as a centrally hosted service with a COBOL onboarding agent or as a local desktop app with a documentation portal. Proprietary, available through Itera.
- [Chain Templar](https://mellicci.dev/work/chain-templar): Decides, at the moment an agent tries to pull in third-party code, whether that agent may acquire it, by checking known-bad intelligence and acquisition risk against a declarative policy, and answering ALLOW · WARN · ASK · DENY. Pre-v1, work in progress.

## Writing

- [Can decision models be used as a prompt injection guard?](https://mellicci.dev/writing/decision-models-prompt-injection-guard): 2026-10-22 — Up to three models screened 10,336 test emails and repository files, mostly from public prompt-injection datasets on Hugging Face and GitHub. Email results look promising; repository results do not yet. An indication from a lab test, not production evidence.
- [New on AI SW Factory: token cost fundamentals for coding agents](https://mellicci.dev/writing/token-cost-fundamentals-course): 2026-10-22 — A free, ten-page module on what a coding agent costs to run and why, built from four mechanics \(re-reading, caching, expiry and generation\) and ending with a controlled cost experiment you can run on your own work.
- [Token cost: is it cheaper to let an agent manage Kubernetes with CLI or MCP?](https://mellicci.dev/writing/cli-or-mcp-kubernetes-agent-token-cost): 2026-10-03 — Four hypotheses tested in 40 measured agent runs across five Kubernetes operations \(incident repair, a failed rollout, a capacity change, a memory fix and a Helm lifecycle\), each evaluated per session with Agent Tokenwatch against independently verified success.