openmark-newsletter-comparison

name	openmark-newsletter-comparison
description	Compose a comparison newsletter from Ahmad's OpenMark saves — A vs B vs C, side-by-side, table-driven. Use when Ahmad wants to compare tools, models, frameworks, approaches, or platforms across his saves. Trigger phrases include "compare", "X vs Y", "head to head", "which should I use", "differences between", "trade-offs between", "/newsletter-comparison". Output centers a markdown table with consistent columns, supports it with short prose blocks, ends with a clear recommendation.
metadata	{"type":"composition"}

OpenMark — Newsletter Comparison

Compare 2-5 things from Ahmad's saves. The table IS the spine. Prose only exists to add context the table can't carry.

When to use vs other newsletter skills

User intent	Skill
"Compare A vs B" / "X vs Y" / trade-offs	this one (comparison)
Single-thesis argument	openmark-newsletter-essay
Categorical weekly recap	openmark-newsletter-roundup
Punchy analytical newsletter	openmark-newsletter

Get the comparison set FIRST

If Ahmad didn't name the items, ASK once:

"Which 2-5 specific things are we comparing? And what's the dimension that matters most — cost, performance, DX, lock-in, license?"

Don't pull data without (a) the items list AND (b) the dimensions. Without both, comparison drifts into recap.

Tool sequence

For each named item:

1. search_semantic(query="<item name>", n=8) — what Ahmad saved about it.
2. find_by_tag(tag="<item-name-lowercase>", n=10) if the item has a tag.

Run these in parallel across items.

Then dedupe across all returned hits. Pick the top 2–3 anchors PER item that ground specific claims (perf numbers, pricing, license terms, DX quotes).

WebFetch the 2 strongest anchors per item to extract specifics. If a fetch fails, drop that anchor.

Output format — NON-NEGOTIABLE

# {Punchy title — name the comparison, ≤ 9 words}

> **{One-sentence recommendation. Who wins for what.}**

{Opening paragraph: 2–3 sentences. Set the stakes — when does this
comparison matter? Who is the reader (someone shipping, evaluating,
choosing)? NO definitions of the items themselves.}

## The table

| Dimension | {Item A} | {Item B} | {Item C} |
|---|---|---|---|
| **License** | MIT | Apache-2.0 | Closed |
| **Cost** | Free | $20/mo seat | Free tier + $99/mo |
| **DX (1-5)** | 4 | 3 | 5 |
| **Lock-in** | Low | Medium | High |
| **Best at** | Speed | Flexibility | Polish |
| **Worst at** | Docs | Latency | Cost |

(Headers and dimensions are EXAMPLES. Use dimensions Ahmad named, plus
any high-signal dimensions the data supports. 4–8 rows. Every cell is
short — phrase, number, or one word.)

## How to read this

{2–4 sentences. Spell out the trade-off pattern the table reveals.
Reference specific cells by name. If two items tie on most rows, name
the one tiebreaker dimension.}

## When to pick each

### Pick **{Item A}** if {one-sentence condition}.

{1–2 sentences of context. Cite the strongest anchor: [phrase](URL).}

### Pick **{Item B}** if {one-sentence condition}.

{Same pattern.}

### Pick **{Item C}** if {one-sentence condition}.

{Same pattern.}

## The numbers (if any)

{Optional section. ONLY include if WebFetch returned concrete numbers.
Format as 2–4 sentences citing the URLs. No fabricated stats.}

## Sources cited

1. [{title}]({URL}) — {which item, what claim it grounds}
2. [{title}]({URL}) — {which item, what claim it grounds}
...

_Comparing {N} items across {M} dimensions · {K} sources cited_

Formatting rules — strictly enforced

• Single H1. No subtitle.
• Recommendation as blockquote under the H1.
• The table is mandatory. 4–8 dimensions, 2–5 items.
• Each cell ≤ 30 chars. If a value needs more, abbreviate or split into a separate row.
• "Pick X if" sections use ### not ## — they're sub-recommendations of the recommendation block.
• Cite URLs inline in the "Pick" sections — [phrase](URL).
• End with ## Sources cited. Required for auto-export.

Voice

Direct. Comparative. No "both are great." Every recommendation names ONE specific use case. Examples:

Generic	Comparison voice
"Both LangChain and LlamaIndex are popular."	"LangChain wins for orchestration. LlamaIndex wins for the retrieval layer."
"It depends on your use case."	"If your team ships features weekly, pick A. If you optimize for cost, pick B."
"There are pros and cons to each."	"C has the best DX and the worst lock-in. Choose your pain."

What NOT to do

• Don't compare more than 5 items. Below 2 is not a comparison; above 5 is a roundup.
• Don't list "pros and cons" in prose. That's what the table is for.
• Don't hedge in the blockquote recommendation. Pick one or name the one tiebreaker.
• Don't include items Ahmad hasn't saved — the comparison must be grounded in his bookmarks.