Generates developer tool feature announcement tweets in an understated, data-driven style. Use when drafting tweets to announce new features, performance improvements, or API additions for projects like Rolldown. Do NOT use for general social media posts, blog writing, or non-technical announcements.
Generate developer tool feature announcement tweets that are understated, proof-driven, and technically credible. Inspired by the style used in Bun's feature announcements.
Before writing anything, build a real understanding of the feature. If the user provides a PR, commit, branch, issue, or file path:
Read the relevant code changes, PR description, or commit messages
Understand what the feature actually does at a technical level
Identify the key insight, what makes this interesting or useful
Look for benchmark data, test results, or before/after comparisons in the codebase
If the user only provides a brief description, ask clarifying questions to fill gaps in understanding. A good tweet requires genuinely understanding the feature. Surface-level descriptions produce surface-level tweets.
Step 2: Classify the Feature Type
Determine which category the feature falls into:
Type
Signal
New API
New function, class, or module being introduced
Performance
Speed, memory, or size improvement with measurable data
CLI feature
New command, flag, or workflow
Compatibility
Standards support, new format, ecosystem integration
Bundled feature
Multiple things combined into one simple experience
Step 3: Gather Input Details
Ask the user for (if not already provided):
Product name (e.g., "Rolldown")
Feature description, what it does
Numbers/benchmarks if performance-related
Contributors to credit (optional)
Upstream projects to credit (optional)
Known limitations or future plans (optional)
Step 4: Draft the Main Tweet
Opening: Always use "In the next version of [Product]"
Body (pick the format matching the feature type):
Feature type
Body format
New API
Function signature in backticks + one-sentence plain-English description
Performance
Short description of what area improved + the key number. Details go in replies
CLI feature
"You can:" + 2-3 escalating CLI examples in backticks
Compatibility
Simple statement: "[Product] gets [X] support"
Bundled feature
Description + punchy tagline ("One file, zero dependencies.")
Rules:
Plain English, minimal jargon
If there is a number, include it -- "2x faster", "reduce objects by 10%"
For performance tweets, explain what improved in terms any web dev would understand. Avoid implementation details like type names, data structures, or internal APIs.
Use "up to X%" instead of ranges like "3-9%". Lead with the peak number.
Don't mention dependency version bumps. Frame improvements as your own work.
Keep it scannable. Use short paragraphs (1-2 sentences) with line breaks between them. A dense block of text is unreadable in tweet form. Put the key number or punchline on its own line.
For performance tweets, lead with the number/result in the main tweet. Move detailed explanation of what changed into an optional reply.
Keep under 280 characters when possible (exclude image placeholders)
End with either a punchy tagline, inline benchmark, or note that an image is attached
Step 5: Draft Thread Replies (Optional)
Thread replies are optional. Only add a reply if it provides genuinely useful context that doesn't fit in the main tweet. A standalone tweet is often better than a forced thread.
If adding replies, each should provide a different useful angle. Pick what fits:
Practical impact or use cases
Context on how it works or why it matters
Proof (benchmarks, test results, before/after data)
Credit to contributors or upstream projects
Roadmap tease or scoping ("This was unrelated to X")
Replies can be longer and more conversational than the main tweet. They're where you explain the "why" and "how" in a natural, human tone. Don't compress them into robotic bullet points.
Rules for replies:
Each reply should stand on its own, a reader might see it without the main tweet
3 replies max, never more
Credit contributors with @mentions, credit upstream projects by name
Step 6: Suggest Visual Proof
For each tweet/reply that would benefit from an image, add a bracketed suggestion:
Scenario
Suggested visual
New API
Polished code screenshot with syntax highlighting and comments explaining usage
Performance claim
Benchmark table or terminal output with before/after/delta
Feature works
Terminal showing command + output
Test rigor
Wall of green checkmarks (pass/fail count)
Comparison
A/B bar charts (red=old, green=new)
CLI feature
Video demo or cheat sheet with usage variations
Principle: the image should be self-contained proof. Someone seeing only the image should understand the claim.
Step 7: Verify Against Anti-Patterns
Check the draft against these, if any match, revise:
No "we're excited to announce..." preamble
No "check out our blog post for more", the tweet IS the content
No vague claims without data, every performance claim has numbers
No superlatives ("blazingly fast", "revolutionary"), the numbers speak
No engagement bait ("what do you think?", "RT if you agree")
No thread longer than 3 replies
No overselling, use understated language ("very slightly faster", "a little")
Numbers are specific and reproducible (test scenario described, hardware noted if relevant)
No em dashes. Use commas, periods, or conjunctions instead.
No "no X, no Y, just Z" constructions (e.g., "No config changes, no API changes. Just faster.")
Don't start sentences with bare past participles ("Optimized...", "Reduced...", "Improved..."). Use "We optimized..." or rephrase.
No awkward specificity ("all 8 benchmarks"). Use natural phrasing ("across our benchmarks").
Don't mention internal dependency names/versions. Frame as your own work.
Be genuine. If the draft reads like an ad or press release, rewrite it.
No walls of text. Use short paragraphs (1-2 sentences) with line breaks. Put the key number on its own line.
Voice Guide
-> See voice-guide for tone rules and examples
Examples
-> See examples for 13 full tweet threads to reference
Output Format
Present the final output as:
**Main tweet:**
[tweet text]
**Visual:** [describe what image/video to create, be specific about content, layout, and what to highlight]
Replies are optional. Only include them if they add genuinely useful context. If including replies, add them after the main tweet:
---
**Reply 1:**
[reply text]
**Visual:** [describe what image/video to create, or "None" if text-only reply]
Visual Description Rules
Every visual suggestion must be actionable. Describe it so the user can go create it:
What to capture: terminal output, code in editor, benchmark run, etc.
What to highlight: which numbers to emphasize, what to annotate (red underlines, "winner" labels, Before/After labels)
Layout: single screenshot, side-by-side comparison, table format, etc.
Format: image (screenshot) vs video (for interactive/workflow demos)
Example visual descriptions:
"Screenshot of terminal running rolldown build showing output. Highlight the bundle size and time in the output."
"Side-by-side polished code screenshot: left shows the config input, right shows the optimized output. Use gradient background with syntax highlighting."
"Before/after benchmark table comparing Rolldown vs esbuild vs webpack. Columns: Bundler, Time, Bundle Size. Bold the Rolldown row."
"Short video demo: run rolldown --watch, edit a file, show the rebuild time in terminal."
1---2name: dev-announce3description: "Generates developer tool feature announcement tweets in an understated, data-driven style. Use when drafting tweets to announce new features, performance improvements, or API additions for projects like Rolldown. Do NOT use for general social media posts, blog writing, or non-technical announcements."4metadata:5 source: https://skillmd.com/skills/bishal-pdmsft/dev-announce6 author: yunfeihe7 version: "1.0.0"8 category: content-creation9 tags: [tweets, announcements, developer-tools]10---1112# dev-announce1314Generate developer tool feature announcement tweets that are understated, proof-driven, and technically credible. Inspired by the style used in Bun's feature announcements.1516## Workflow1718Copy this checklist and check off each step:1920```21- [ ] Step 1: Understand the feature (read code/PR/commits if available)22- [ ] Step 2: Classify the feature type23- [ ] Step 3: Gather input details24- [ ] Step 4: Draft main tweet25- [ ] Step 5: Draft thread replies (optional)26- [ ] Step 6: Suggest visual proof27- [ ] Step 7: Verify against anti-patterns28```2930### Step 1: Understand the Feature3132Before writing anything, build a real understanding of the feature. If the user provides a PR, commit, branch, issue, or file path:3334- Read the relevant code changes, PR description, or commit messages35- Understand what the feature actually does at a technical level36- Identify the key insight, what makes this interesting or useful37- Look for benchmark data, test results, or before/after comparisons in the codebase3839If the user only provides a brief description, ask clarifying questions to fill gaps in understanding. A good tweet requires genuinely understanding the feature. Surface-level descriptions produce surface-level tweets.4041### Step 2: Classify the Feature Type4243Determine which category the feature falls into:4445| Type | Signal |46|---|---|47| **New API** | New function, class, or module being introduced |48| **Performance** | Speed, memory, or size improvement with measurable data |49| **CLI feature** | New command, flag, or workflow |50| **Compatibility** | Standards support, new format, ecosystem integration |51| **Bundled feature** | Multiple things combined into one simple experience |5253### Step 3: Gather Input Details5455Ask the user for (if not already provided):56- Product name (e.g., "Rolldown")57- Feature description, what it does58- Numbers/benchmarks if performance-related59- Contributors to credit (optional)60- Upstream projects to credit (optional)61- Known limitations or future plans (optional)6263### Step 4: Draft the Main Tweet6465**Opening:** Always use "In the next version of [Product]"6667**Body** (pick the format matching the feature type):6869| Feature type | Body format |70|---|---|71| New API | Function signature in backticks + one-sentence plain-English description |72| Performance | Short description of what area improved + the key number. Details go in replies |73| CLI feature | "You can:" + 2-3 escalating CLI examples in backticks |74| Compatibility | Simple statement: "[Product] gets [X] support" |75| Bundled feature | Description + punchy tagline ("One file, zero dependencies.") |7677**Rules:**78- Plain English, minimal jargon79- If there is a number, include it -- "2x faster", "reduce objects by 10%"80- For performance tweets, explain what improved in terms any web dev would understand. Avoid implementation details like type names, data structures, or internal APIs.81- Use "up to X%" instead of ranges like "3-9%". Lead with the peak number.82- Don't mention dependency version bumps. Frame improvements as your own work.83- Keep it scannable. Use short paragraphs (1-2 sentences) with line breaks between them. A dense block of text is unreadable in tweet form. Put the key number or punchline on its own line.84- For performance tweets, lead with the number/result in the main tweet. Move detailed explanation of what changed into an optional reply.85- Keep under 280 characters when possible (exclude image placeholders)86- End with either a punchy tagline, inline benchmark, or note that an image is attached8788### Step 5: Draft Thread Replies (Optional)8990Thread replies are optional. Only add a reply if it provides genuinely useful context that doesn't fit in the main tweet. A standalone tweet is often better than a forced thread.9192If adding replies, each should provide a different useful angle. Pick what fits:93- Practical impact or use cases94- Context on how it works or why it matters95- Proof (benchmarks, test results, before/after data)96- Credit to contributors or upstream projects97- Roadmap tease or scoping ("This was unrelated to X")9899Replies can be longer and more conversational than the main tweet. They're where you explain the "why" and "how" in a natural, human tone. Don't compress them into robotic bullet points.100101**Rules for replies:**102- Each reply should stand on its own, a reader might see it without the main tweet103- 3 replies max, never more104- Credit contributors with @mentions, credit upstream projects by name105106### Step 6: Suggest Visual Proof107108For each tweet/reply that would benefit from an image, add a bracketed suggestion:109110| Scenario | Suggested visual |111|---|---|112| New API | Polished code screenshot with syntax highlighting and comments explaining usage |113| Performance claim | Benchmark table or terminal output with before/after/delta |114| Feature works | Terminal showing command + output |115| Test rigor | Wall of green checkmarks (pass/fail count) |116| Comparison | A/B bar charts (red=old, green=new) |117| CLI feature | Video demo or cheat sheet with usage variations |118119Principle: the image should be self-contained proof. Someone seeing only the image should understand the claim.120121### Step 7: Verify Against Anti-Patterns122123Check the draft against these, if any match, revise:124125- [ ] No "we're excited to announce..." preamble126- [ ] No "check out our blog post for more", the tweet IS the content127- [ ] No vague claims without data, every performance claim has numbers128- [ ] No superlatives ("blazingly fast", "revolutionary"), the numbers speak129- [ ] No engagement bait ("what do you think?", "RT if you agree")130- [ ] No thread longer than 3 replies131- [ ] No overselling, use understated language ("very slightly faster", "a little")132- [ ] Numbers are specific and reproducible (test scenario described, hardware noted if relevant)133- [ ] No em dashes. Use commas, periods, or conjunctions instead.134- [ ] No "no X, no Y, just Z" constructions (e.g., "No config changes, no API changes. Just faster.")135- [ ] Don't start sentences with bare past participles ("Optimized...", "Reduced...", "Improved..."). Use "We optimized..." or rephrase.136- [ ] No awkward specificity ("all 8 benchmarks"). Use natural phrasing ("across our benchmarks").137- [ ] Don't mention internal dependency names/versions. Frame as your own work.138- [ ] Be genuine. If the draft reads like an ad or press release, rewrite it.139- [ ] No walls of text. Use short paragraphs (1-2 sentences) with line breaks. Put the key number on its own line.140141## Voice Guide142143-> See [voice-guide](reference/voice-guide.md) for tone rules and examples144145## Examples146147-> See [examples](reference/examples.md) for 13 full tweet threads to reference148149## Output Format150151Present the final output as:152153```154**Main tweet:**155156[tweet text]157158**Visual:** [describe what image/video to create, be specific about content, layout, and what to highlight]159```160161Replies are optional. Only include them if they add genuinely useful context. If including replies, add them after the main tweet:162163```164---165166**Reply 1:**167168[reply text]169170**Visual:** [describe what image/video to create, or "None" if text-only reply]171```172173### Visual Description Rules174175Every visual suggestion must be actionable. Describe it so the user can go create it:176177- **What to capture:** terminal output, code in editor, benchmark run, etc.178- **What to highlight:** which numbers to emphasize, what to annotate (red underlines, "winner" labels, Before/After labels)179- **Layout:** single screenshot, side-by-side comparison, table format, etc.180- **Style:** plain terminal screenshot vs polished code screenshot (gradient background, syntax highlighting)181- **Format:** image (screenshot) vs video (for interactive/workflow demos)182183Example visual descriptions:184- "Screenshot of terminal running `rolldown build` showing output. Highlight the bundle size and time in the output."185- "Side-by-side polished code screenshot: left shows the config input, right shows the optimized output. Use gradient background with syntax highlighting."186- "Before/after benchmark table comparing Rolldown vs esbuild vs webpack. Columns: Bundler, Time, Bundle Size. Bold the Rolldown row."187- "Short video demo: run `rolldown --watch`, edit a file, show the rebuild time in terminal."
Run npx get-skill add bishal-pdmsft/dev-announce in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Generates developer tool feature announcement tweets in an understated, data-driven style. Use when drafting tweets to announce new features, performance improvements, or API additions for projects like Rolldown. Do NOT use for general social media posts, blog writing, or non-technical announcements. It is listed under Integrations & APIs on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
bishal-pdMSFT (@bishal-pdmsft) published this skill. Their other Agent Skills are listed on their SkillMD profile.