Technical Briefing: skillsz

The Problem

AI agents need reusable, focused workflows for common tasks like codebase improvement, research, and documentation. Without packaged skills, agents repeatedly rediscover the same patterns or rely on generic prompts that lack the specificity required for reliable outcomes. This repository addresses that gap by organizing David Ondrej's operational patterns into loadable agent skills.

What This Does

The repository packages 51 focused workflows into SKILL.md files across five category folders under skills/. Each skill explains when and how to use it for repeatable agentic work.

Category structure:

  • skills/agent-orchestration/ — 14 skills for running, scheduling, and coordinating AI coding agents, including agent-self-scheduling, goal-loop, handoff, and herdr (herd routing)
  • skills/skill-authoring/ — 5 skills for creating, improving, and distributing agent skills and context files
  • skills/research-and-web/ — 10 skills for web research, browser harnessing, and API data pulling, including browser-harness, deep-research, and pi-web-search
  • skills/thinking-and-docs/ — 16 skills for structured thinking, interviewing, teaching, and documentation transformation, including brain-to-docs, decisions, and teach/ format guides
  • skills/ops-and-setup/ — 5 skills for machine/Server setup, security configuration, and operations, including anti-sleep (shell script), vps-server-management, and google-safe-browsing

Three hook scripts live in hooks/: deny-dangerous.sh, test-guard.sh, and dangerous-patterns.txt — guardrails for agent operations.

How It Is Wired

The import graph resolves to 0 internal modules and 0 import edges across 3 analyzed shell files (hooks/deny-dangerous.sh, hooks/test-guard.sh, and one additional shell file). This is a skill-collection repository, not a modular codebase; value derives from loading individual SKILL.md files rather than interconnected functions.

Entry points are the SKILL.md files themselves. Each documents its own trigger conditions and usage patterns. The hook scripts (hooks/test-guard.sh, hooks/deny-dangerous.sh) provide boundary enforcement for agent operations but do not participate in a call graph — they are invoked externally as guard conditions.

No file touches a database, network endpoint, or external filesystem beyond what individual skills describe. The hooks/ scripts operate on local file patterns and process substitution.

How To Use It

Setup: No build step or package installation is required. The repository is a static collection of Markdown files and shell scripts. Clone with:

git clone https://github.com/moses-y/skillsz

Configuration: No environment variables or config files are required. Skills are self-contained Markdown documents. The LICENSE file (MIT) governs reuse.

Running it: Load a skill by referencing its SKILL.md path. Example workflows include:

  • Loading skills/agent-orchestration/goal-loop/SKILL.md to enable agent self-improvement loops
  • Using skills/research-and-web/deep-research/SKILL.md for multi-step research tasks
  • Invoking hooks/test-guard.sh as a pre-flight check before agent deployment

No package.json, pyproject.toml, or Dockerfile exists. Operation is purely file-based.

Real-World Use

A research agent investigating a new topic can load skills/research-and-web/pi-web-search/SKILL.md to standardize web search patterns, then switch to skills/research-and-web/browser-harness/SKILL.md when interactive browser automation is required. Similarly, a code review agent can load skills/agent-orchestration/codex-subagent/SKILL.md to invoke specialized sub-agent behavior instead of relying on generic instructions.

The hook scripts provide a consistent gate: running hooks/test-guard.sh before agent deployment validates that dangerous patterns are not present in the target codebase.

Code Health & Issues

Measured findings from static analysis:

  • No CI/CD pipeline detected — no automated build/test gate (/.github/ or CI config absent) [Medium/SDLC]
  • 2 test files present but no CI to run them
  • No Dockerfile present
  • LICENSE present (MIT)
  • No lockfile present
  • No committed secrets found

SDLC observations: The repository lacks automated testing gates and CI configuration. This is appropriate for a skill-collection repository where each skill is a standalone Markdown document, but teams intending to integrate these skills into agent pipelines should establish their own test and integration gates.

The Bottom Line

This is a well-organized collection of 51 focused agent skills spanning orchestration, research, and operations. The skills are practical and self-documenting, making them easy to load and deploy in agent workflows. The absence of CI/CD and test infrastructure is expected for this format but should be supplemented by the consuming team. Best suited for teams building agentic systems who need reusable, focused workflow primitives rather than a turnkey framework.