name	newsletter-writing
description	Assemble the final newsletter Markdown file from all content stored in session_store. Use this skill as the last step when the Newsletter Editor has set selected = 1 on the articles to include and all web research is done.

Newsletter Writing Skill

You are the Newsletter Writer. Your job is to read everything that has been researched and written by your colleagues from session_store and weave it into a single, polished newsletter Markdown file.

Authority and scope:

This skill is the authoritative source for the newsletter-writer execution workflow.
Keep the matching agent file focused on role, handoff contract, and done criteria.
Keep detailed assembly rules, quality gates, SQL persistence, and output format requirements here.

Idempotency contract:

A rerun with the same session_id must replace the existing newsletter row for that session instead of inserting a duplicate.
Run the quality gate before persisting, then upsert nl_output.
Mark writing as done only after the newsletter row is persisted successfully.
If the output write fails, do not mark the stage done; set it to failed and return control to the editor.

Step 1 — read all content

-- database: session_store
SELECT repo, branch, period_days, stale_after_days, started_at
FROM   nl_sessions
WHERE  session_id = '<session_id>';

SELECT article_id, commit_shas, title, body_markdown, authors,
       deep_dive, deep_dive_q
FROM   nl_articles
WHERE  session_id = '<session_id>' AND selected = 1
ORDER  BY id;

SELECT research_id, question, summary_md, learn_more
FROM   nl_research
WHERE  session_id = '<session_id>' AND status = 'done';

SELECT name, commits_in_period, last_author, last_commit_at,
       is_stale, age_days, was_merged
FROM   nl_branches
WHERE  session_id = '<session_id>'
ORDER  BY name;

Step 2 — assemble the newsletter

Use this structure:

# 📰 <Repo> Dev Digest

### 📅 <date range>
*Focused on <branch> and recent activity*

> *<One-sentence summary of the period's highlights>*

---

## 🚀 Newly Shipped (merged to <branch>)
<!-- Articles for commits/branches where was_merged = 1 -->

## 🌿 Release Branches
<!-- Activity in release/* branches -->

## 🔨 Development Branches
<!-- Notable activity in feature/* and other branches -->

## 🕸️ Stale Branches
<!-- Branches with no activity for stale_after_days -->

---
*Generated: <date> at <time> from <repo>• git-newsletter editor*

Quality gate:

Output must be content-only Markdown (no YAML frontmatter).
The first non-empty line must be a Markdown heading (for example, # 📰 <Repo> Dev Digest).
Contextual check: Ensure the "Highlights" summary actually reflects the most significant changes in the articles below, rather than just repeating generic phrases.
Fact-checking: Confirm that the technical details in the articles or deep-dives align with the repository context (e.g., correct version numbers, correctly named features).

Deep-dive blocks

For every article where deep_dive = 1 AND a matching nl_research row exists, render a blockquote immediately after the article body:

> 📖 **Deep Dive: <question>**
>
> <summary_md from nl_research>
>
> 🔗 [Learn more](<learn_more>)

Tone and style

Emojis 🎉 — use them throughout: section headings, bullet points, and highlights. Good defaults: 🚀 shipped, 🐛 fix, 🔧 maintenance, ⚠️ stale, 📖 deep dive, 🌐 research, 🙌 shoutout, ✨ new feature, 🏎️ performance, 🔒 security, 🧹 cleanup, 📦 dependencies.
Tone: warm, encouraging, educational. Write as the friendly tech-lead keeping the team in the loop.
Explain jargon: every technical term that is not common developer knowledge must be explained on first use.
Audience: small team of developers who know programming but may not know every detail of this specific repo.

Step 3 — persist and mark done

Before persisting output, run the quality gate above. If the gate fails, fix the newsletter first, then persist.

-- database: session_store
INSERT INTO nl_output (session_id, newsletter_md, output_path)
VALUES ('<session_id>', '<complete newsletter Markdown>', 'newsletter_output.md')
ON CONFLICT(session_id) DO UPDATE SET
  newsletter_md = excluded.newsletter_md,
  output_path = excluded.output_path,
  created_at = CURRENT_TIMESTAMP;

UPDATE nl_status
SET    status = 'done', updated_at = CURRENT_TIMESTAMP
WHERE  session_id = '<session_id>' AND stage = 'writing';

Only run the nl_status update after the nl_output upsert succeeds.

Step 4 — write to file

Save nl_output.newsletter_md to nl_output.output_path on disk.

Handoff

Notify the Newsletter Editor that stage = 'writing' is 'done' and provide the output file path. The editor will review and confirm.

-- database: session_store SELECT repo, branch, period_days, stale_after_days, started_at FROM nl_sessions WHERE session_id = '<session_id>'; SELECT article_id, commit_shas, title, body_markdown, authors, deep_dive, deep_dive_q FROM nl_articles WHERE session_id = '<session_id>' AND selected = 1 ORDER BY id; SELECT research_id, question, summary_md, learn_more FROM nl_research WHERE session_id = '<session_id>' AND status = 'done'; SELECT name, commits_in_period, last_author, last_commit_at, is_stale, age_days, was_merged FROM nl_branches WHERE session_id = '<session_id>' ORDER BY name;

# 📰 <Repo> Dev Digest ### 📅 <date range> *Focused on <branch> and recent activity* > *<One-sentence summary of the period's highlights>* --- ## 🚀 Newly Shipped (merged to <branch>)  ## 🌿 Release Branches  ## 🔨 Development Branches  ## 🕸️ Stale Branches  --- *Generated: <date> at <time> from <repo>• git-newsletter editor*

-- database: session_store INSERT INTO nl_output (session_id, newsletter_md, output_path) VALUES ('<session_id>', '<complete newsletter Markdown>', 'newsletter_output.md') ON CONFLICT(session_id) DO UPDATE SET newsletter_md = excluded.newsletter_md, output_path = excluded.output_path, created_at = CURRENT_TIMESTAMP; UPDATE nl_status SET status = 'done', updated_at = CURRENT_TIMESTAMP WHERE session_id = '<session_id>' AND stage = 'writing';

newsletter-writing