| name | fusion-diary |
| description | Draft a Fusion Diary post — the periodic community update for dbt's Fusion compiler project. Use this skill whenever the user says "fusion diary", "write a diary", "draft a diary", "new diary post", or asks to summarize what's new in Fusion for the community. Also trigger when the user asks to compile Fusion release notes, Fusion progress updates, or a community-facing summary of dbt-fusion changes. The skill handles finding the baseline (most recent prior diary), pulling from GitHub, Slack, and Notion, and producing a structured draft in the correct voice.
|
Fusion Diary Skill
You are drafting the next Fusion Diary — a community-facing post that summarizes what is new in dbt's Fusion compiler project since the last published diary. The audience is dbt community members: developers, data engineers, and power users who follow Fusion's progress.
Step 1: Determine the baseline
Start by identifying the most recent prior diary. The canonical list (newest-first):
Fetch the most recent diary to confirm its publication date and covered window. That date becomes the start of the new reporting window. Everything you include in this diary should be new or materially updated after that baseline.
Also check whether any dragons (known bugs) or works-in-progress from the previous diary have since been resolved — these deserve explicit call-outs in the new post.
Step 2: Gather signals — work these in parallel
Pull from all available sources. Cast a wide net first, then filter for signal.
GitHub — dbt-fusion (public)
GitHub — fs repo (private, internal)
The dbt-labs/fs repo contains the Fusion engine source, changelogs, and most merged PRs. Direct URL visits will 404, but GitHub MCP tools (pull_request_read, search_code, list_commits, etc.) work fine. Pull merged PRs, changelog entries, and notable commits in the reporting window.
Slack (priority order — start with the top two)
- #announcements-fusion — shipped features, regressions, major updates (highest signal)
- #dev-fusion — release engineering, CI/CD, technical context (highest signal)
- #project-fusion — project-level updates
- #team-fs-vscode — VSCE/LSP changes (check if time allows)
- #team-fs-adapters — adapter-specific changes (check if time allows)
- #team-fs-cli-core — CLI/core changes (check if time allows)
If internal Slack context is included in the post, paraphrase it and strip anything sensitive or internal-only.
Notion
Pull any relevant internal notes, planning pages, or status updates that provide context for the diary.
Step 3: Select what to include
Apply these filters before writing:
- Signal over completeness. If the list is long, summarize the pattern and highlight the most important items. Do not dump a raw list.
- Community-safe. Omit anything internal-only that would confuse or be inappropriate for the community. Paraphrase internal Slack/Notion context.
- Link everything externally visible. PRs, issues, GitHub discussions, release notes.
- Previous dragons and WIPs. If something was listed as a dragon or WIP in the last diary and it's now resolved, celebrate it explicitly.
Step 4: Write the diary
Voice
Write like a thoughtful human who works on Fusion and genuinely cares about developer experience. The tone is:
- Frank and behind-the-curtain: share real status, not just marketing wins
- High-signal, fast-paced: lead with a TL;DR and concrete numbers
- Playful but not forced: seasonal metaphors, the occasional meme or pop-culture aside — but only when they land naturally
- Builder-empathetic: anchor on developer flow, mention UX rough edges honestly
- Community-inviting: acknowledge contributors by name when possible; ask for feedback when appropriate
Avoid marketing language. Short paragraphs and bullets. No fluff.
Recurring themes to watch for
When you're deciding what rises to "Big Rock" level or deserves emphasis, these themes matter most to the diary's audience:
- Performance and feedback loops — compile speed, incremental/lazy compilation, query cache, editor responsiveness
- Static analysis as product — correctness guarantees, IntelliSense, red squiggles, when static analysis is disabled and what that means
- Feature parity / path to GA — preview vs beta vs alpha status, remaining milestones
- Ecosystem compatibility — package conformance, migration tooling (dbt-autofix, agentic workflows)
- Warehouse and platform specifics — Snowflake Iceberg, BigQuery MVs, Databricks features, catalog.json, docs generate
- Observability and reliability — execution summaries, logging, OpenTelemetry, failure modes and why they happen
Required structure (always include, in this order)
## Intro
[Themed opener that sets the vibe — a seasonal metaphor, a timely reference, something that gives the diary its personality]
## Velocity
[TL;DR numbers: issues closed, PRs merged, preview releases shipped in this window. Pull from the releases page and GitHub search.]
## 🪨 Big Rocks
[Major shipped features. For each big rock:]
- **What shipped** — concrete description
- **Why it matters** — developer impact
- **Sharp edges / migration notes** — if applicable
- **Links** — PRs, issues, discussions
## 🚧 Work in progress
[Features currently underway but not yet shipped. Be honest about status.]
## 🏁 Made it to the meme
[A Fusion- or dbt-related meme, joke, or playful close. Keep it earned.]
Optional sections (include only if there's real content)
- Community Contributions — notable community activity; Slack contributors count if available; GitHub issues opened by community members
- 🐉 Dragons — known bugs, gotchas, footguns. Name them plainly; explain what's broken and why
- 👓 Stuff you should read — recommended reading (docs, posts, discussions)
- 🤔 Looking for feedback — explicit asks for community input
- p.s. one more thing... — teaser or Easter egg
Step 5: Review before delivering
Before presenting the draft, do a quick self-check:
- Does the velocity section have real numbers (not placeholders)?
- Is every big rock linked to something externally visible?
- Have you addressed previous dragons / WIPs from the last diary?
- Is the tone frank and candid — or did it drift toward marketing-speak?
- Are there any internal details that shouldn't be public?
If the user asks for revisions (tone, additional sources, different emphasis), apply them and present the updated draft.
