NeoLabHQ/context-engineering-kit

NeoLabHQ / Context‑Engineering‑Kit

What it is – A collection of ready‑made plugins (called “skills” in the agentskills.io spec) that you can add to LLM‑powered coding assistants such as Claude Code, Gemini CLI, Antigravity, Cursor, OpenCode, etc. Each plugin bundles a small set of commands, agents and context snippets designed to make the assistant more reliable, token‑efficient and focused on a concrete development practice (e.g., reflexion, spec‑driven development, code review, test‑driven development, domain‑driven design, etc.).

Why it matters – Large language models tend to lose accuracy as the prompt (context) grows. The kit solves this by:

• loading only the agents you need, keeping the token budget low;
• injecting proven “context‑engineering” patterns (reflection loops, sub‑agents acting as judges, specification‑first prompts) that have been measured to raise the probability of correct output from ~60 % to >90 % on real‑world code changes;
• exposing those patterns as simple slash‑commands (/reflect, /memorize, /do‑and‑judge, …) that can be invoked from the IDE or CI pipelines.

Key components

Plugin	Core idea	Typical command(s)
Reflexion	Self‑refinement loop – ask the model to critique and improve its own answer.	/reflect, /memorize, /critique
Spec‑Driven Development (SDD)	Generate a formal spec first, then let the model implement it; includes a built‑in code‑reviewer sub‑agent.	/plan‑task, /implement‑task, /brainstorm
Subagent‑Driven Development (SADD)	Spins up a fresh sub‑agent for each step and runs a judge between steps to catch regressions.	/do‑and‑judge, /do‑in‑steps
Review	Multi‑agent code‑review (bug‑hunter, security‑auditor, etc.) with impact/confidence filtering.	/review‑local‑changes, /review‑pr, /triage‑review
Git	Convenience wrappers for commits, PR creation, issue analysis, worktree handling.	/commit, /create‑pr, /analyze‑issue
TDD, DDD, Tech‑Stack, Kaizen, FPF, Docs, MCP	Domain‑specific helpers that inject language‑specific best‑practices, test scaffolding, architectural guidelines, continuous‑improvement techniques, etc.

Installation – The kit is distributed as a marketplace that can be added to the target agent:

# Claude Code (per‑plugin selection)
/plugin marketplace add NeoLabHQ/context-engineering-kit
/plugin install reflexion@NeoLabHQ/context-engineering-kit

Other environments (Gemini CLI, Antigravity, Vercel‑labs skills, OpenSkills) install the whole bundle with a single command, after which you can delete unwanted skills manually.

Typical workflow

1. Ask the model to perform a task, e.g. claude "implement user authentication".
2. Trigger a reflexion hook (/reflect) either manually or by appending “reflect” to the prompt.
3. The model analyses its own output, fixes obvious issues, and optionally stores learned fixes with /memorize for future tasks.
4. For larger changes you can switch to /do‑and‑judge or the full SDD flow, which adds a specification step, a step‑wise implementation, and a final review.
5. Use the Git plugin to commit, open a PR, or load issue data, and optionally run the Review plugin in a GitHub Action for automated PR checks.

Who should use it

• Developers who already rely on LLM‑based coding assistants and want more predictable, higher‑quality output.
• Team leads / DevOps looking to embed AI‑driven code generation into CI pipelines with minimal token cost.
• Researchers / Tool builders interested in concrete, open‑standard implementations of context‑engineering patterns (agentskills.io, Arc42, self‑refine, reflexion, etc.).

Where to learn more – Full docs, plugin reference and installation guides live at the public GitBook site: https://neolab.gitbook.io/cek/.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.