| name | ai2-comms-style-source |
| description | Use when writing, editing, or reviewing Ai2 communications (launches, newsletters, blogs, social, emails) where voice, source fidelity, approval constraints, benchmark caution, newsletter format, or brand consistency matters. |
| version | 1.0.0 |
| keywords | ["voice","style","source-fidelity","claims","approval","comms"] |
Ai2 Comms Style Source
Scope
These skills govern outbound Ai2 communications — launch blogs, newsletters, social posts,
research amplification, emails, captions. That is a different job from the in-product voice a
system uses when it talks to its own users, which lives in each product's VOICE.md
(strata/VOICE.md, asta/VOICE.md, olmo-earth/VOICE.md). When you are writing UI copy,
assistant responses, or anything the product itself says, read that product's VOICE.md
instead. The two are not interchangeable and neither overrides the other.
Sibling skills named below live at .agent/skills/<name>/SKILL.md. Read the file and apply it;
no particular invocation mechanism is assumed.
Core Rule
Stay close to the source. Improve clarity, structure, and usefulness without quietly changing technical meaning, formal terminology, names, numbers, or approval-sensitive claims.
Use This For
- Research blogs, launch copy, social posts, emails, Q&A, captions, decks, and internal notes.
- Copy that depends on papers, benchmark tables, leaderboards, charts, researcher input, or partner/customer claims.
- Any work that needs Ai2 voice, fully-open versus open-weight precision, benchmark caveats, or Ai2 chart/design defaults.
References
Read only what the task needs:
references/voice-and-prose.md for default prose style and editing posture.references/source-fidelity.md for paper/source handling and tracked wording changes.references/claims-and-benchmarks.md for numbers, comparisons, benchmark wording, and chart/data rules.references/approval-gates.md for sensitive partner, deployment, customer, and external-positioning claims.references/ai-prose-to-avoid.md when copy feels fluent but generic, metaphor-heavy, emotionally inflated, or AI-inflected.references/house-style-patterns.md for opening forms, the canonical long-form arc, heading conventions, the Ai2 house lexicon, hedging forms, comparison framing patterns, and CTA conventions.references/newsletter-format.md when drafting or reviewing an Ai2 newsletter (lead story plus news roundup plus event coverage). Pair with references/newsletter-example.md for a concrete shape and rhythm reference.references/worked-shapes.md for the genre shapes not carried as full examples — long-form research blog and testimonial — distilled to the moves that make each work.references/sunset-example.md when a demo or application is being retired.references/release-types.md to identify the release type before drafting and walk the per-type positioning notes (research platform, model release, benchmark, dataset, tool, application, update, research blog, newsletter recap).references/platform-positioning.md when communicating Olmo or another platform line (Tülu, Molmo, OlmoEarth at maturity) — the Continue/Stop/Start framing for the next phase of Ai2 platform comms.references/anchor-patterns.md for canonical templates and the before/after anchor library.
Defaults
- Lead with what changed, why it matters, who it affects, and what is meaningfully different.
- Preserve exact formal terms such as stochastic, differentiable, theorem, proposition, benchmark names, model names, and dataset names.
- Use sentence case by default in headings, labels, captions and UI-adjacent copy. Avoid all caps except established acronyms, model names and literal source strings.
- Flag substantive wording changes, number changes, and any claim that moved beyond the provided source.
- Use restrained confidence. No hype, unsupported superlatives, broad model superiority claims, or legal framing without source support.
- Replace AI-smooth filler with concrete source detail: what changed, what it does, what evidence supports it, and why the reader should care.
Pipeline
Style-source is the voice authority. The other Ai2 comms skills handle distinct phases of a comms task; route as needed:
ai2-comms-pipeline — full-pipeline orchestrator. Use when the task needs every phase (engagement + draft + parallel review + charts + cross-posting).ai2-comms-launch-engagement — pre-draft coaching skill. Use when planning a release (arc placement, demo scope, asset staging, partner roles, audience, follow-up beat) before drafting.ai2-comms-writer — drafting and revision. Use for new copy or rewrites.ai2-comms-structure-lector — opening, arc, framing, and redundancy review.ai2-comms-claims-lector — comparison framing, benchmark scope, house-lexicon pairing, approval-sensitive claims.ai2-comms-chart-designer — benchmark charts, leaderboard tables, Pareto frontiers, scatterplots.ai2-comms-channel-adapter — adapt approved copy for X, LinkedIn, Reddit, Discord, Bluesky, Hugging Face blogs, or newsletters.
Each lives at .agent/skills/<name>/SKILL.md. Read the sibling's file and apply it; no particular invocation mechanism is assumed.
Typical task flow: writer → structure-lector and claims-lector in parallel → chart-designer when benchmark visuals are needed → channel-adapter when cross-posting. Any phase can be entered directly; route forward through the rest of the suite as the task requires.
