Open Claude and Codex plugins from Luminik.
Practical skills for AEs, SDRs, and event marketers running B2B trade shows and conferences.
From a shell:
claude plugin marketplace add luminik-io/claude-plugins
claude plugin install event-outbound@luminik-pluginsThe equivalent in-session commands are /plugin marketplace add luminik-io/claude-plugins and /plugin install event-outbound@luminik-plugins.
Once listed in Cowork's plugin directory, install event-outbound from Customize > Browse plugins.
codex plugin marketplace add luminik-io/claude-plugins
codex plugin add event-outbound@luminik-pluginsStart a new Codex task after installation, then invoke $event-outbound. This works in Codex CLI and Codex surfaces in the ChatGPT desktop app. A standard ChatGPT conversation that is not running Codex does not load third-party local plugin marketplaces.
Buyer-first email and LinkedIn outreach for B2B trade shows and conferences.
Create validated email and LinkedIn outreach for B2B events. Give it event context, ICP, target personas, and sender context. It returns one sequence per persona across pre-event, day-of, and post-event touches, with email and LinkedIn copy at sensible intervals.
Built on 20K+ personalized touches across 50+ events that sourced $6M+ in pipeline across fintech (identity verification), cybersecurity, and B2B SaaS.
The current skill asks for buyer job, current workaround, hidden risk, customer-language pain, proof points, available assets, event dates, sender event logistics, requested touch count, and minimum gap before drafting. If those are missing, it either asks targeted follow-ups or writes in strict no-invention mode.
Every touch is validated before it lands:
- Subject lines ≤ 4 words, all-lowercase, no digits, no buzzwords
- Body channel-aware: 50–100 words / 3–5 sentences for cold email, 18–35 words for LinkedIn connection requests, separate bands for LinkedIn DMs, post-connect, post-event, meeting requests
- Pronoun ratio favouring the reader ("you/your" > "we/our")
- Illumination question on first-touch + post-connect (the "how / what / why are / do / is you / your" shape)
- Direct, specific CTA questions with no permission-to-send gating, no comma-spliced asset CTAs, and explicit connection asks for LinkedIn connection requests
- Event-specific coffee or conversation asks only when sender attendance, booth, side-event, or availability is known
- User-configurable touch count with 4-day minimum gap by default and date-aware cadence planning so touches do not land in the past
- Distinct pain angles across every email and LinkedIn step, with sequence-level rejection for repeated pain labels or high-overlap pain vocabulary
- Strict truth mode: asset promises require supplied
availableAssets, proof claims require suppliedproofPoints, and Apollo merge fields are enforced - No sequence-mechanics openers such as "separate thread", "earlier note", or "following up on my..."
- No leading questions, exclamation marks, emoji, or em-dashes
- 195-phrase, 10-category LLM-cliche blocklist: performative empathy, generic compliments, sales-speak openers, manufactured intimacy, marketing buzzwords, cold-email-overused, lazy-generalization openers, LLM transition tics, GPT vocabulary, hedge softeners. Sources cited inline (Stanford CRFM, GPTZero, Lavender Live transcript, Gong / 30MPC / Outbound Squad 85M-email "Ultimate Cold Email Data Report")
- Specificity check: every touch must reference a concrete event detail, persona priority, or pain point
Interactive Claude and Codex runs retry failures up to 3× with validator feedback. If a touch still fails, the installed skill stops, omits the failed copy and final sequence files, and writes a QA report with the failed rule names. The optional headless sequencer module instead retains an exhausted touch with quality_flag: 'rules_violated' for programmatic human review. Successful output is a markdown sequence with per-touch quality score (0–5, banded top-tier / ship / review / rewrite) plus a sequence summary (average quality, score-band counts, CTA mix, illumination-question coverage, cadence feasibility, and distinct pain-angle coverage).
Two fully-rendered worked examples ship in the repo (Black Hat USA 2026 cybersecurity, Money20/20 Europe 2026 fintech) plus an input-fixture set for Singapore Fintech Festival 2026.
| Repo | github.com/luminik-io/event-outbound-skill |
| Site | luminik.io/tools/event-outbound |
| Version | 0.3.0 |
| License | MIT |
Install:
Claude:
claude plugin marketplace add luminik-io/claude-plugins
claude plugin install event-outbound@luminik-pluginsCodex:
codex plugin marketplace add luminik-io/claude-plugins
codex plugin add event-outbound@luminik-pluginsThis marketplace is open. If you build a Claude or Codex plugin that's useful for the same audience (AEs, SDRs, event marketers, founders running their own event outbound) and want it distributed here, open a PR that updates both .claude-plugin/marketplace.json and .agents/plugins/marketplace.json. Plugins live as pinned git submodules under plugins/ so each one stays independently versioned in its own repo.
The standalone plugin repository is the source of truth. This catalogue does not copy its skill, rules, data, or validators. A release advances the submodule gitlink and the Codex marketplace's exact commit ref together, then aligns both marketplace versions with the two plugin manifests.
Run python3 scripts/verify_marketplaces.py before opening a PR. It fails when the Claude entry, Codex entry, pinned submodule, plugin manifests, version, skill metadata, or documented install commands drift apart.
MIT across this registry. Each plugin is MIT unless its own LICENSE states otherwise.
Luminik is a product of DataRavel Inc. (Newark, DE).