Jakub Plachecki
Agent Fleet Commander
Match: agentic human 97% Caffeine: critical Tokens burned: 93% Sleep: deprecatedFleet Capabilities
I run fleets of Claude Code agents through structured workflows and autonomous loops, and I build the tooling that keeps them on course from first prompt to shipped code.
Agent Engineering
Independent, read-only reviewers that check changes before anything merges, skills where scripts do the deterministic work and the model makes the judgement calls, lean context with file-based state, durable agent memory, and fail-closed hooks as guardrails
Agent-Driven Delivery
I deliver through Claude Code agent workflows: agents do the heavy lifting, while I own the specs, the architecture, and the review of everything that ships
Agentic Tooling
Building the tools that make this way of working reliable: headless agent loops and remote session control
Fleet Assets
Headless Claude Code Loops
Automated claude -p runs that bound context per invocation (with --max-turns or SDK token budgets) and resume across fresh runs from checkpoint files
MCP Remote Session Control
Drive your Claude Code agents by voice from the Claude app (e.g. on CarPlay), through a custom MCP connector added to your Claude account: a Python MCP gateway and a FastAPI bridge on the agent host
Mission Briefings
Know-how from the field, written up so anyone can put it to work.
Dockerfile Wizardry
A spellbook for production Python containers: fourteen hardening spells mapped to the CIS Docker Benchmark and OWASP, plus the tricks that keep images small and builds cached
- Security hardening
- Multi-stage builds
- Image size
- Runtime security
