
Luca Dev Kit
- Updated July 23, 2026
- lucadellanna/luca-kit
luca-dev-kit automates developer workflow: pre-PR quality gates, an autonomous Gemini review loop, and pre-commit hook management. A developer uses it to enforce checks and get automated cross-model review before opening a PR.
Key points
- Pre-PR quality gates
- Autonomous Gemini review loop
- Pre-commit hook management
Luca Dev Kit by the numbers
- Data as of Jul 24, 2026 (Skillselion catalog sync)
/plugin marketplace add lucadellanna/luca-kit/plugin install luca-dev-kit@lucadellannaAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Last updated | July 23, 2026 |
|---|---|
| Repository | lucadellanna/luca-kit ↗ |
What it does
Developer workflow automation: pre-PR quality gates, autonomous Gemini review loop, and pre-commit hooks.
README.md
Luca's plugins
Status: Early access / Pre-release. These plugins are under active development. Features may change, break, or be removed without notice. Not ready for production use.
This marketplace ships four Claude plugins:
| Plugin | Audience | Purpose |
|---|---|---|
| luca-kit | Anyone | Skill discovery, self-reflection, CLAUDE.md auditing, and productivity hooks |
| luca-reflection-kit | Anyone | Self-reflection and cross-session learning: scan conversations for improvement points and mine recurring patterns |
| luca-ops-kit | Organizations and individuals | Craft reusable skills, maintain a self-improving setup, and govern your Claude skill library |
| luca-dev-kit | Developers | Pre-PR quality gates, autonomous Gemini review loop, and pre-commit hook management |
Most companies and people using AI are stuck at the "clever individual prompts" stage: useful experiments, inconsistent execution, little reuse, and no lasting memory. These plugins provide guided workflows to make good procedures explicit and reusable, so know-how doesn't stay trapped in individual heads or chat histories.
Installation
Install in Claude Code, Cowork, or both; if you use both apps, follow each procedure separately.
luca-kit
Claude Code
Requires Claude Code.
/plugin marketplace add lucadellanna/luca-kit/plugin install luca-kit@lucadellanna- Enable auto-updates: type
/plugin, press Tab twice to open the Marketplaces tab, selectluca-kit, and select Enable updates.
Claude Cowork
- Left sidebar → Customize → Plugins → Personal → + → Add marketplace →
lucadellanna/luca-kit - Click + next to
luca-kit - Enable auto-updates: Browse Plugins → Personal → luca-kit → ··· → Sync automatically
To uninstall:
/plugin uninstall luca-kit@lucadellanna/luca-kit
luca-ops-kit
Claude Code
Requires Claude Code.
/plugin marketplace add lucadellanna/luca-kit/plugin install luca-ops-kit@lucadellanna- Enable auto-updates: type
/plugin, press Tab twice to open the Marketplaces tab, selectluca-ops-kit, and select Enable updates. - Run
/luca-ops-recommended-setupto add best-practice rules and see a privacy/backup checklist.
Claude Cowork
- Left sidebar → Customize → Plugins → Personal → + → Add marketplace →
lucadellanna/luca-kit - You will see all available luca-kit plugins; click + next to each one you want to install
- Enable auto-updates: Browse Plugins → Personal → luca-ops-kit → ··· → Sync automatically
- Run
/luca-ops-recommended-setupto add best-practice rules and see a privacy/backup checklist.
To uninstall: If you ran /luca-ops-recommended-setup, run /undo-setup first to remove rules it added to your global Claude config. Then:
/plugin uninstall luca-ops-kit@lucadellanna/luca-kit
luca-dev-kit
Claude Code
Requires GitHub CLI (gh) and Gemini Code Assist installed on the repo.
/plugin marketplace add lucadellanna/luca-kit/plugin install luca-dev-kit@lucadellanna- Enable auto-updates: type
/plugin, press Tab twice to open the Marketplaces tab, selectluca-dev-kit, and select Enable updates. - Run
/luca-dev-recommended-setupto install pre-commit hooks and review the Gemini Code Assist requirement forreview-loop(persistent local checkout only, not needed in Conductor).
To uninstall:
/plugin uninstall luca-dev-kit@lucadellanna/luca-kit
luca-reflection-kit
Claude Code
Requires Claude Code.
/plugin marketplace add lucadellanna/luca-kit/plugin install luca-reflection-kit@lucadellanna- Enable auto-updates: type
/plugin, press Tab twice to open the Marketplaces tab, selectluca-reflection-kit, and select Enable updates. - Run
/luca-reflection-kit:luca-reflection-recommended-setupto configure session notes and privacy settings.
Claude Cowork
- Left sidebar → Customize → Plugins → Personal → + → Add marketplace →
lucadellanna/luca-kit - You will see all available luca-kit plugins; click + next to each one you want to install
- Enable auto-updates: Browse Plugins → Personal → luca-reflection-kit → ··· → Sync automatically
- Run
/luca-reflection-kit:luca-reflection-recommended-setupto configure session notes and privacy settings.
To uninstall:
/plugin uninstall luca-reflection-kit@lucadellanna/luca-kit
Skills
luca-ops-kit
| Skill | What it does |
|---|---|
| luca-ops-recommended-setup | First-run wizard: adds three best-practice rules to your Claude config |
| undo-setup | Reverses everything /luca-ops-recommended-setup added so you can cleanly uninstall the plugin |
| build-work-context | Interviews you about your company and role, then saves a persistent profile so Claude doesn't need to ask "who do you work for?" every session |
| create-skill | Turns a procedure, SOP, checklist, or verbal description into a ready-to-use skill file, scoring and improving it before saving |
| audit-skill | Scores a single skill against 7 quality dimensions (clarity, security, instruction explicitness, and more), proposes improvements, and iterates until the bar is met |
| audit-skills | Scans your whole skill library for overlapping skills, then audits a rotating batch of 3 so every skill gets reviewed over time (requires luca-kit) |
luca-kit
| Skill | What it does |
|---|---|
| list-skills | Lists every installed skill across all plugins with its plugin, one-line description, and line count in a single table |
| reflect | Highlights what to improve in how you and Claude work together, so the next conversation goes better |
| compact-claude-files | Tightens your CLAUDE.md, memory, and path-rule files with within-file compressions and a loss-verifier safety net |
| restructure-claude-files | Moves content out of CLAUDE.md to memory, surfaces cross-type transfer candidates, and flags scope mismatches between project and global |
luca-dev-kit
| Skill | Trigger | What it does |
|---|---|---|
| luca-dev-recommended-setup | /luca-dev-recommended-setup |
One-time setup: installs pre-commit hooks, notes Gemini Code Assist requirement. Run once per machine in persistent local checkouts. |
| open-pr | "open pr", "create pr", "/open-pr" | Triple review, fix findings, typecheck, push, create PR, hand off to review-loop |
| review-loop | Auto-invoked by open-pr; or "review loop" | Polls Gemini, classifies threads, applies fixes, re-triggers review, repeats until clean |
| triple-review | "triple review" | Three-lens parallel review: principles, recurring bug patterns, structural integrity |
| targeted-review | "/targeted-review", "targeted review", "focused review of " | Single-file ad-hoc review with a derived (or user-supplied) checklist; bug-only findings via a structural FINDINGS marker. Catches bugs broad reviews miss |
| specs-adherence-review | "check specs", "adheres to principles?" | Checks changed code against CLAUDE.md rules |
| install-pre-commit-hooks | Invoked by luca-dev-recommended-setup | One-time hook setup: em-dash check, gitleaks, optional typecheck |
Hooks
luca-ops-kit
Installed automatically with the plugin (no setup needed):
| Hook | Event | What it does |
|---|---|---|
| claude-md-tidy | PostToolUse (Edit/Write) | After any edit to a CLAUDE.md or AGENTS.md file, injects a 7-criterion review (conciseness, duplication, contradictions, scope, ephemerality, vague triggers, context pollution) with quantitative metrics |
| stop-apology-check | Stop | When the response contains a self-correction phrase (e.g. "you're right", "my mistake", "I missed") without a ★ rule-update widget, rule-file edit, error-log append, or explicit one-off escape, blocks the stop and asks Claude to apply the Error → rule update contract. Forces meta-cognitive learning to actually happen instead of being deferred. See plugins/luca-ops-kit/CLAUDE.md for the full contract. |
| hedge-scan | PostToolUse (Edit/Write/MultiEdit) | On edits to rule-like files (CLAUDE.md, SKILL.md, hook scripts, command files), warns when added list-item lines contain hedge words ("try to", "consider", "prefer", "should probably"). Strips quoted spans first, so hedge words used as quoted examples are not flagged. Enforces the "rules must use imperative language" principle. |
luca-reflection-kit
| Skill | What it does |
|---|---|
| reflect | After a session, extracts what went well, what went wrong, and what should become a memory update, skill improvement, or new skill |
| dream | Mines your /reflect logs to surface patterns across sessions: recurring issues never fixed, memory contradictions, and improvements that keep coming up but never land |
| setup-context-search | (Power users) Installs qmd and configures it as an MCP server; enables semantic overlap detection in /reflect and /dream. Requires Node.js 22+. |
Hooks
| Hook | Event | What it does |
|---|---|---|
| consolidate-reflect-queue | PreToolUse | Before /reflect runs in a Conductor workspace, merges queued learnings from sibling workspace folders into the current workspace queue |
| optimization-hint | UserPromptSubmit | On every prompt, reminds Claude to append a one-sentence Optimization hint at the end of the current response if it involves 8+ tool calls (memory-worthy pattern / skill to edit or improve) |
| workflow-hint | UserPromptSubmit | On every prompt, reminds Claude to append a one-sentence Workflow hint at the end of the current response if it involves 8+ tool calls (skill to codify / user judgement or workflow to automate / friction to remove) |
How it works
luca-ops-kit
The plugin ships meta-skills: structured workflows for building, auditing, and improving procedures. You can use it standalone to create, audit, and govern your own skill library.
Who uses it:
| Role | What they do with it |
|---|---|
| Manager | Points it at a procedure and asks what should become a skill |
| Admin | Turns a repeated reporting task into a reusable workflow |
| Salesperson | Converts a successful call-prep process into a skill |
| Frontline manager | Builds an SOP from tacit know-how |
| Local power user | Audits, consolidates, and governs the team's skill library |
luca-dev-kit
Write "open pr" and Claude handles the rest: pre-PR quality gates (triple-review against principles, recurring bug patterns, and structural integrity), PR creation, and an autonomous Gemini review loop that fixes comments and re-triggers review until the PR is clean.
luca-reflection-kit
Run /reflect after a session to extract what went well, what didn't, and what should become a memory update, skill improvement, or new skill. Run /dream periodically to mine past reflect logs for recurring patterns and improvements that keep coming up but never land.
Context search (optional, power users)
As your file library grows (notes, SOPs, context files), finding the right document becomes harder. /setup-context-search installs qmd, a local semantic search engine, and wires it into Claude as an MCP server. After setup, /reflect and /dream use semantic search to detect conceptual overlaps when proposing new rules or skills, and Claude can search your files natively in any session.
Requirements: Node.js 22+, npm or bun, ~2.2 GB disk for local AI models (runs entirely on your machine, no cloud).
What it enables:
| Query type | Example | How it works |
|---|---|---|
| Keyword | "Find the Acme Corp file" | Fast BM25 full-text search |
| Semantic | "Notes similar to X" | Vector similarity via local embeddings |
| Hybrid | "Recent project notes on topic Y" | Keyword + semantic + LLM re-ranking |
Run /luca-reflection-kit:setup-context-search once; Claude gets the search tools permanently.
For holdco customers
If your organization receives luca-ops-kit through a holding company, franchisor, or trade association, skills are delivered in layers. Your holdco curates domain skills (the actual business procedures for your industry) on top of luca-ops-kit's meta-skills, and your team can further adapt them to local context.
| Layer | Who provides it | What it contains |
|---|---|---|
| luca-ops-kit | Luca Dellanna | Meta-skills: build, audit, and improve procedures |
| Holdco layer | Your holding company, franchisor, or trade association | Domain skills tailored to your industry (e.g., onboarding checklists, compliance workflows, reporting templates) |
| Partner layer | Your team | Local adaptations: company-specific terminology, approval chains, and tooling integrations |
Disclaimer
These plugins (luca-ops-kit, luca-dev-kit, and luca-reflection-kit) are for business use only, and are intended exclusively for businesses and professionals acting in the course of a trade or profession. By using them, you represent that you are not a consumer under Italian law (Article 3, Codice del Consumo). They are designed exclusively for use with Claude (Anthropic); any adaptation to other AI providers is unsupported and at the user's sole risk. They are provided "as-is", without warranty of any kind, express or implied.
The plugins depend on third-party services (including Anthropic (Claude API) and Claude Code) that Luca Dellanna does not control. Changes to those services, including model deprecations, pricing changes, or discontinuation, are not Luca Dellanna's responsibility. The plugins run locally on your machine; Luca Dellanna does not collect or process your data.
Outputs are generated by Anthropic Claude; the plugins orchestrate prompts but do not themselves produce content. AI-generated outputs may contain errors, fabrications, or hallucinations, and may appear authoritative but be factually wrong. You are solely responsible for reviewing all outputs before acting on them. These plugins are not a substitute for professional legal, financial, medical, or other expert advice; do not use them for legal filings, medical decisions, financial transactions, regulated activities, or high-risk AI applications (EU AI Act, Reg. 2024/1689) without independent expert review.
To the maximum extent permitted by applicable law, and in any case limited to fees paid in the 12 months preceding the claim, Luca Dellanna accepts no liability for any loss, damage, or consequence, whether direct or indirect, arising from use of these plugins or reliance on their outputs. This includes losses caused by third-party service changes (Anthropic, Claude Code, etc.). Nothing in this disclaimer excludes liability that cannot be excluded under applicable law (including liability for fraud, gross negligence, or death or personal injury caused by negligence). Full terms will be published at luca-dellanna.com before commercial launch. This disclaimer is governed by Italian law; any disputes shall be resolved in the Italian courts.
Disclaimer v1.0, 2026-05-10
License
Source-available, not open-source. You may inspect the repository and use it for personal evaluation. Commercial, client-facing, team, or organizational use requires a paid license. Contact Luca@Luca-Dellanna.com for more information.