nbbaier/agent-skills

Purpose and Distribution

This repository is a portable collection of instruction-driven “skills” for coding agents, not an executable application or service. Each skill packages a primary agent workflow in a SKILL.md, optionally supported by reference material and static assets. Consumers install or copy selected skill directories into an agent’s skill location.

Plugin metadata is provided at .claude-plugin/plugin.json; OpenCode-oriented metadata/configuration lives in .oc/config.json and .oc/config.md.

Primary Entry Points

The repository-level catalog and installation path are defined in README.md. Agent runtimes discover individual capabilities through their skill instruction documents, especially:

• skills/blueprint/SKILL.md

• skills/goal-refiner/SKILL.md

• skills/implementation-guide/SKILL.md

There is no server, build pipeline, database, or application runtime represented in the tree.

Skill Package Boundary

Each skills/<name>/ directory is an independently consumable capability boundary. Its SKILL.md is the authoritative operational prompt/workflow; colocated references/ files provide deeper procedures, templates, or domain guidance without necessarily being loaded first.

Representative packages include:

• Planning/specification: skills/blueprint/SKILL.md, skills/blueprint/references/plan-template.md

• Design discovery: skills/ideation/SKILL.md, skills/goal-refiner/SKILL.md

• Implementation/review guidance: skills/implementation-guide/SKILL.md, skills/code-refactor-review/SKILL.md

Planning and Execution Flow

The core planning-oriented flow is: refine an ambiguous goal, investigate and shape options, produce a decision-level specification, then generate execution-ready implementation guidance. blueprint explicitly contains supporting stages for reconnaissance, decomposition, destinations, and validation/closure:

• skills/blueprint/references/recon.md

• skills/blueprint/references/decomposition.md

• skills/blueprint/references/closing-the-loop.md

These documents make the repository particularly suited to downstream agent planning: they separate problem framing and evidence gathering from plan decomposition and completion criteria.

Specialized Guidance Modules

Several standalone skills constrain or improve particular engineering activities rather than orchestrating a full delivery lifecycle:

• UI/prototyping: skills/wireframe/SKILL.md, skills/html-tools/SKILL.md

• Frontend correctness: skills/react-effect-patterns/SKILL.md

• Operations/observability: skills/logging-best-practices/SKILL.md

• Visualization critique: skills/tufte-viz/SKILL.md

• Knowledge validation: skills/quiz-gate/SKILL.md

wireframe is the notable asset-bearing package, including a reusable static viewer template at skills/wireframe/assets/site-template.html.

External Dependencies and Lifecycle

The repository relies primarily on agent-host runtimes that understand installed skill directories, with explicit integration signals for Claude plugins and OpenCode. Installation may use Vercel’s external skills CLI (npx skills add ...), but no package manifest or runtime dependency is present here.

Legacy capabilities are intentionally isolated under skills/deprecated/, including a historical npm supply-chain audit shell script at skills/deprecated/npm-supply-chain-security/scripts/audit.sh. They should not be treated as part of the active architecture unless explicitly revived.