Agent skills

Skills that outlive subscriptions.

Every AI vendor wants to lock your workflows into their runtime. We weave the opposite: brand-agnostic, cross-compatible agent skills — protocols and tools that run identically wherever your agents live. Your process shouldn't be a hostage of your subscription.

Works with Claude CodeCodex CLIAntigravityOpen harnesses

The pattern book

The skills line.

Open source, in production use, and installable today.

Sprint Loops

Loop engineering protocol

The problem: long-horizon autonomous coding drifts — agents lose the plot across sessions, redo work, and leave no audit trail.
The skill: a lightweight five-phase protocol (Research → Plan → Build → Test → Loop) with persistent state on disk. The filesystem is the state machine. Ships three ways: a Claude Code plugin marketplace, a Codex CLI skill, and a runtime-agnostic spec for open harnesses.

GECK

Structured memory framework

The problem: context amnesia. Every CLI-agent session starts from zero — repeated mistakes, forgotten decisions, no coherence across a large project.
The skill: the Garden of Eden Creation Kit — a protocol plus generator that gives agents typed tasks, decision records, learnings, and episodic logs in a structured folder, read within a declared context budget.

MDR

Download resilience skill

The problem: multi-hour, multi-gigabyte model downloads from HuggingFace die at 87% and nobody notices until morning.
The skill: Massive Download Resilience — procurement of large models with nested cron jobs that watch, resume, and verify until the weights are truly on disk. Works from Claude Code, Codex, and Antigravity.

Supporting cast: oovra (compose and diff agentic system prompts from standardized .md + TOML elements) and .lux (a document format designed for human and AI readers as first-class consumers).

The commission

Two ways to work with us.

01

Custom Skill & Protocol Development

Your team has a workflow — code review, release management, data pipeline babysitting — that an agent should run reliably. We design and build it as a portable skill: documented, versioned, and working across the harnesses you already use. One protocol, every runtime.

Scope a skill

02

Agentic Development Workshops

Hands-on training for engineering teams: how to actually work with Claude Code, Codex CLI, and local harnesses — loop engineering, context management, permission models, and the habits that separate "tried the demo" from "shipped with agents." Remote or on-site, sized to your team.

Book a workshop

Questions first? Email charles@threadandsignal.com.