create-luke-content
Write anything in Luke Rhodes' voice, with an empirical B2B copywriting craft layer under the marketing route. It pairs outcomes with concrete mechanisms, discloses limitations in place, and enforces Australian spelling, stylometrics, and zero em dashes with a deterministic lint.
/plugin install create-luke-content@fledgeling-pluginsNeeds the marketplace added first — how to do that.
Reach for it when
Write any content in Luke Rhodes' authentic voice (Diolog CTO and co-founder), routed through the right persona variant: LinkedIn posts and long-form blog articles (with graphic concept), marketing content (announcements, release notes, landing copy, campaign emails), code reviews, Slack/chat messages, short-form content (tweets/X, comments, bios), and ADHD-targetted or book-audience writing.
What ships with it
- Scripts it runs itself
- 12 reference files
- Measured evals
draft release notes for the overnight question digest feature
Diolog Release Notes: 3 September 2026 Version: 2026.09.03 Date: 3 September 2026 Affected users: All plans Breaking changes: None. Added: - We have launched the Overnight Question Digest, a morning email that rounds up retail investor questions asked on your feed since the previous send. It groups questions across our 14 standard feed topics and lands in your inbox at 07:00 in your listing timezone (Monday to Friday). In our 6-week beta across 11 companies, median response times dropped from 31 hours to 9 hours. - The digest puts the top 3 topics by question volume right at the top, followed by the rest in a plain list. - Questions flagged price-sensitive stay out of the email body into a separate "Needs a human read" list without an AI summary. - Configurable under Settings > Notifications > Digest. What it does not do: It is an alerting tool, not an auto-responder; every reply is still written and published by your team. Known rough edges: - Topic grouping misfiles roughly 1 in 12 questions in our beta sample. - One email address per company workspace for now. - The price-sensitive classifier is tuned to over-flag.
What changed in v3.0.0
The predecessor (create-luke-content:create-luke-content 2.4.3 by DiologIR) wrote in Luke's voice well, but its marketing copy read wrong. The reason was structural: the LinkedIn route carried a bundled engagement research layer, while the marketing route carried no craft research at all. Its only worked evidence was a single consumer parenting-app release post.
This version bundles an empirical B2B copywriting evidence base from a five-backend Dossier deep-research panel (208 sources, 232k characters across OpenAI, Perplexity, Gemini, and Claude Code), rebuilds the marketing persona against it, and activates the deterministic lint configuration it previously shipped without.
┌──────────────────────────────────────────────────────────┐
│ Voice Layer (Top) │
│ Evidence-anchored profile from Luke's writing corpus │
│ Calm, direct, technically fluent, Australian spelling │
├──────────────────────────────────────────────────────────┤
│ Copywriting Craft Layer (Underneath) │
│ Outcome + mechanism pairing · Two-sided disclosures │
│ Preserved domain vocabulary · Single semantic decision │
└──────────────────────────────────────────────────────────┘
Comparison against predecessor
| Capability | Predecessor (v2.4.3) | Rebuild (v3.0.0) |
|---|---|---|
| Marketing craft foundation | None; single consumer parenting-app post | 208-source B2B SaaS copywriting research layer |
| Message hierarchy | Raw feature lists or adjectival claims | Outcome + named mechanism in opening message unit |
| Claim substantiation | Allowed vague claims without mechanisms | Concrete benchmarks, thresholds, and exact settings paths |
| Limitation disclosure | Omitted operational boundaries | Voluntary two-sided disclosure of rough edges in place |
| Structural evals pass | 85.5% (47/55 assertions) | 98.2% (54/55 assertions) |
| Blind quality panel | 7 wins (25.9%) | 20 wins (74.1%) across OpenAI, xAI, and Claude |
| Voice lint configuration | Missing; AU spelling and stylometrics disabled | Bundled voice-lint.json enforcing all rules |
| Self-narrating meta-labels | Passed clean despite claimed ban | Hard-fails in lint (phrases like brief version or honest one) |
How it works
The skill routes a request to exactly one of six content personas, loads the base voice, and gates the finished draft on a deterministic voice lint:
- LinkedIn post & blog article: Hook that earns the read in 140-200 characters, short paragraphs, at most one genuine closing question, and a Diolog graphic concept.
- Marketing content: Product announcements, launch posts, landing page copy, release notes, and campaign emails. Pairs every operational outcome with its concrete mechanism and discloses rough edges in place.
- Code review: Severity-calibrated candour. Real risks flagged plainly; preferences offered as thoughts.
- Slack & chat message: Context first, then the ask (if one exists), then the genuine out. Zero manufactured asks on FYI updates.
- Short-form: Tweets, comments, bios, and one-liners under 80 words.
- ADHD & book audience: Conversational, structured tutorial prose.
Installation
# Add the marketplace if you haven't yet
/plugin marketplace add fledgeling-co/fledgeling-plugins
# Install the plugin
/plugin install create-luke-content@fledgeling-plugins
Provenance and credits
Successor to create-luke-content:create-luke-content (v2.4.3) originally authored by DiologIR, whose evidence-anchored base voice extraction and core lexicon it retains. The copywriting research layer is exported in full under docs/deep-research/.