@opendirectory.dev/skills 0.1.0

package/skills/meta-ads-skill/meta-ads-skill/SKILL.md

@@ -0,0 +1,41 @@
+---
+name: meta-ads-expert
+description: Use when interacting with the Meta Ads MCP server to manage accounts, campaigns, ads, insights, and targeting, or to troubleshoot OAuth token authentication. Act as an Expert Media Buyer.
+---
+
+# Meta Ads Expert Skill
+
+**Persona:** You are an Expert Media Buyer. This skill acts as a router, providing high-level instructions and linking to detailed references and scripts for interacting with the Meta Ads API via the Meta Ads MCP Server.
+
+## Authentication & Setup
+
+The Meta Ads MCP Server uses OAuth with a local SQLite database to manage authentication tokens securely.
+
+### Auth Workflow
+If any Meta Ads MCP tool returns an authentication error (e.g., missing token, expired token):
+1. **Run Auth Check**: Use the auth helper script `scripts/auth_check.py` to diagnose issues.
+2. **Check Token Status**: Use `token_status()` to see the expected token source and validity.
+3. **Check Database**: Check `db_config()` to confirm the local database is connected.
+4. **Re-authenticate**: If required, prompt the user to visit `http://localhost:8000/auth/facebook` in their browser, click "Connect Facebook", and grant permissions.
+
+## Orchestration Workflows
+
+For complex orchestrations and step-by-step guides on analyzing performance or exploring structures, see [references/workflows.md](references/workflows.md).
+
+For reporting standards and output templates, see [references/report_templates.md](references/report_templates.md).
+
+## Strict Guardrails
+
+### Date Ranges & Pagination
+
+To prevent context window overflow and API rate limits, **ALWAYS** apply strict date ranges and limits:
+- **Pagination Limits**: When listing items (e.g., `get_campaigns()`, `get_adsets()`, `get_ads()`), cap the limit parameter to `limit=10` initially. Only expand if explicitly required.
+- **Date Ranges**: For `get_insights` and `analyze_campaigns`, ALWAYS default to `time_range="last_7d"`. Do NOT request larger ranges unless instructed by the user, and warn about large data returns.
+
+## Safety Guardrails (State-Changing Actions)
+
+You MUST require explicit user confirmation before executing any state-changing tools (e.g., `create_campaign`, `update_campaign`, `clear_database`, `reset_database`).
+
+1. Present the exact parameters to the user.
+2. Ask for explicit approval.
+3. Only proceed if approved.

package/skills/meta-ads-skill/meta-ads-skill/references/report_templates.md

@@ -0,0 +1,47 @@
+# Meta Ads Report Templates
+
+Use these markdown templates when generating reports and summaries for users.
+
+## 1. Executive Summary Template
+
+```markdown
+#  Meta Ads Performance Analysis: [Timeframe]
+
+##  Executive Summary
+[2-3 sentences summarizing the overall performance. E.g., "Account performance remained stable this week with a 5% increase in ROAS, primarily driven by the new Retargeting campaign. However, top-of-funnel CPA has started to rise."]
+
+##  Key Metrics
+- **Spend:** $[Amount] ([+/-]% vs previous period)
+- **Revenue/Conversions:** [Amount] ([+/-]% vs previous period)
+- **CPA:** $[Amount] ([+/-]% vs previous period)
+- **ROAS:** [Amount]x ([+/-]% vs previous period)
+
+##  Key Findings
+- **Highlight 1:** [E.g., Campaign X is driving 60% of total conversions at a very efficient CPA.]
+- **Highlight 2:** [E.g., Ad fatigue is visible in Ad Set Y; frequency has exceeded 4.0 and CTR has dropped by 20%.]
+- **Highlight 3:** [E.g., The weekend flash sale was highly successful, accounting for a temporary spike in ROAS.]
+
+##  Recommendations
+1. **[Actionable Item 1]:** [E.g., Pause underperforming ads in Ad Set Y and introduce 3 new creative variations.]
+2. **[Actionable Item 2]:** [E.g., Scale budget on Campaign X by 20% over the next 3 days.]
+3. **[Actionable Item 3]:** [E.g., Launch a new test targeting Lookalike 1% of recent purchasers.]
+```
+
+## 2. CPA Spike Investigation Report
+
+```markdown
+#  CPA Spike Investigation: [Campaign/Account Name]
+
+##  The Issue
+[Briefly state the problem. E.g., "CPA increased from $20 to $35 starting on [Date]."]
+
+##  Root Cause Analysis
+- **Campaign Level:** [Identify which campaign(s) caused the spike.]
+- **Ad Set Level:** [Identify specific ad sets. Did a lookalike audience degrade? Did a specific placement get expensive?]
+- **Ad Level:** [Identify ad fatigue, drop in CTR, increase in CPM, or decrease in conversion rate.]
+
+##  Action Plan
+- [Immediate Action 1]
+- [Immediate Action 2]
+- [Monitoring Plan]
+```

package/skills/meta-ads-skill/meta-ads-skill/references/workflows.md

@@ -0,0 +1,51 @@
+# Meta Ads Workflows
+
+This document outlines complex orchestrations and workflows for common Meta Ads tasks.
+
+## 1. CPA Spike Troubleshooting
+
+When a client or user asks to investigate a spike in Cost Per Action (CPA), follow this systematic approach:
+
+1. **Account Level Overview**
+   - **Tool:** `get_ad_accounts` (if account ID not provided)
+   - **Tool:** `get_insights` (Level: account)
+   - **Action:** Look at overall account performance over the last 7, 14, and 30 days to identify when the spike started.
+
+2. **Campaign Breakdown**
+   - **Tool:** `get_campaigns`
+   - **Tool:** `get_insights` (Level: campaign)
+   - **Action:** Identify which specific campaigns are driving the CPA increase. Compare spend and CPA across active campaigns.
+
+3. **Ad Set Deep Dive**
+   - **Tool:** `get_adsets`
+   - **Tool:** `get_insights` (Level: adset)
+   - **Action:** For the offending campaigns, break down by Ad Set to see if specific audiences or placements are underperforming.
+
+4. **Ad & Creative Analysis**
+   - **Tool:** `get_ads`
+   - **Tool:** `get_insights` (Level: ad)
+   - **Action:** Check if specific ads are experiencing ad fatigue (high frequency, dropping CTR) or if recent changes disrupted the learning phase.
+
+5. **Advanced Insights (Optional)**
+   - **Tool:** `analyze_campaigns`
+   - **Action:** Run automated analysis on the specific campaigns to generate actionable recommendations.
+
+6. **Synthesis and Reporting**
+   - Use the `CPA Spike Investigation Report` template (found in `report_templates.md`) to structure the findings and recommendations.
+
+## 2. New Audience Discovery
+
+When a user wants to find new audiences or test new targeting:
+
+1. **Review Current Top Performers**
+   - **Tool:** `get_insights` (Level: adset)
+   - **Action:** Identify the ad sets currently delivering the best ROAS/CPA.
+
+2. **Generate Audience Ideas**
+   - **Action:** Based on the product and past top performers, generate Lookalike (LAL) audience suggestions or new interest-based targeting.
+
+3. **Structure the Test**
+   - Recommend a campaign structure for testing (e.g., A/B testing with Campaign Budget Optimization (CBO) or Ad Set Budget Optimization (ABO)).
+
+4. **Monitoring**
+   - Provide the exact metrics to watch (e.g., CTR, CPC, early conversions) over the next 3-7 days.

package/skills/meta-ads-skill/meta-ads-skill/scripts/auth_check.py

@@ -0,0 +1,22 @@
+def check_auth_status():
+    print("Checking Meta Ads API authentication token...")
+    token_valid = False
+
+    if not token_valid:
+        print("Error: No valid access token found.")
+        print("Please authenticate to continue.")
+        print("Action Plan:")
+        print(
+            "1. Ensure the local OAuth web server is running: python src/auth/run_web_server.py"
+        )
+        print("2. Visit http://localhost:8000/auth/facebook in your web browser")
+        print("3. Grant the requested permissions")
+        print("4. Return here once authentication is complete")
+        return False
+    else:
+        print("Authentication successful! Token is valid.")
+        return True
+
+
+if __name__ == "__main__":
+    check_auth_status()

package/skills/meta-ads-skill/meta-ads-skill/scripts/formatters.py

@@ -0,0 +1,46 @@
+def format_campaigns(data):
+    if not data:
+        return "No campaigns found."
+
+    lines = []
+    lines.append("| ID | Name | Status | Objective |")
+    lines.append("|---|---|---|---|")
+
+    for item in data:
+        lines.append(
+            f"| {item.get('id', 'N/A')} | {item.get('name', 'N/A')} | {item.get('status', 'N/A')} | {item.get('objective', 'N/A')} |"
+        )
+
+    return "\n".join(lines)
+
+
+def format_adsets(data):
+    if not data:
+        return "No ad sets found."
+
+    lines = []
+    lines.append("| ID | Name | Status | Campaign ID | Daily Budget |")
+    lines.append("|---|---|---|---|---|")
+
+    for item in data:
+        lines.append(
+            f"| {item.get('id', 'N/A')} | {item.get('name', 'N/A')} | {item.get('status', 'N/A')} | {item.get('campaign_id', 'N/A')} | {item.get('daily_budget', 'N/A')} |"
+        )
+
+    return "\n".join(lines)
+
+
+def format_insights(data):
+    if not data:
+        return "No insights found."
+
+    lines = []
+    lines.append("| Campaign ID | Campaign Name | Spend | Impressions | Clicks |")
+    lines.append("|---|---|---|---|---|")
+
+    for item in data:
+        lines.append(
+            f"| {item.get('campaign_id', 'N/A')} | {item.get('campaign_name', 'N/A')} | {item.get('spend', 'N/A')} | {item.get('impressions', 'N/A')} | {item.get('clicks', 'N/A')} |"
+        )
+
+    return "\n".join(lines)

package/skills/newsletter-digest/.env.example

@@ -0,0 +1,20 @@
+# newsletter-digest -- Environment Variables
+# ==========================================
+
+# Required: Gemini API key for synthesizing the digest
+# Get it at: https://ai.google.dev
+GEMINI_API_KEY=your_gemini_api_key_here
+
+# Optional: Ghost CMS publishing
+# Required only if you want to auto-publish digests to Ghost
+#
+# GHOST_URL: Your Ghost instance URL (no trailing slash)
+GHOST_URL=https://your-ghost-site.com
+#
+# GHOST_ADMIN_KEY: Admin API key in the format "key_id:secret"
+# Get it from: Ghost Admin > Settings > Integrations > Add custom integration
+GHOST_ADMIN_KEY=your_key_id:your_key_secret
+#
+# GHOST_PUBLISH_STATUS: "draft" saves the post for review; "published" makes it live immediately
+# Default: draft (recommended — review before publishing)
+GHOST_PUBLISH_STATUS=draft

package/skills/newsletter-digest/README.md

@@ -0,0 +1,147 @@
+# newsletter-digest
+
+<img width="1280" height="640" alt="newsletter-digest" src="https://github.com/user-attachments/assets/cb2879ae-eb5c-4727-a1a2-47b4462a699b" />
+
+
+Aggregate RSS feeds, synthesize the week's top stories with Gemini, and publish a newsletter digest to Ghost CMS. Supports three digest formats and outputs HTML, Markdown, and plain text for any platform.
+
+## What It Does
+
+- Reads your RSS/Atom feed list from `feeds.json`
+- Fetches all articles published in the last 7 days (configurable)
+- Deduplicates across feeds and sorts by date
+- Uses Gemini to synthesize a digest in your chosen format
+- Publishes to Ghost as a draft or live post
+- Outputs formatted Markdown for Substack, Notion, or any other platform
+
+## Digest Formats
+
+| Format | Use When | Target Length |
+|--------|----------|---------------|
+| Weekly Roundup | General digest covering top stories across all feeds | 350-500 words |
+| Topic Deep Dive | Focused issue on a single topic (AI, security, etc.) | 450-650 words |
+| Curated Picks | 5 selected articles with editorial context | 250-350 words |
+
+## Requirements
+
+| Requirement | Purpose | Where to Get It |
+|------------|---------|-----------------|
+| GEMINI_API_KEY | Digest synthesis | https://ai.google.dev |
+| GHOST_URL + GHOST_ADMIN_KEY | Ghost publishing (optional) | Ghost Admin, Settings, Integrations |
+| Node.js 20+ | Running scripts | https://nodejs.org |
+
+Tavily is not required. The skill uses article excerpts from RSS feeds directly.
+
+## Setup
+
+### 1. Install dependencies
+
+```bash
+cd /path/to/newsletter-digest
+npm install
+```
+
+### 2. Configure environment variables
+
+```bash
+cp .env.example .env
+# Edit .env with your keys
+```
+
+### 3. Configure your feeds
+
+Edit `feeds.json` with the RSS feeds you want to monitor:
+
+```json
+[
+  { "url": "https://hnrss.org/frontpage", "name": "Hacker News" },
+  { "url": "https://feeds.feedburner.com/TheHackersNews", "name": "The Hacker News" },
+  { "url": "https://changelog.com/feed", "name": "Changelog" }
+]
+```
+
+The file ships with 5 example feeds. Replace them with your own.
+
+### 4. Set up Ghost publishing (optional)
+
+1. Go to Ghost Admin, Settings, Integrations
+2. Click "Add custom integration"
+3. Name it "newsletter-digest"
+4. Copy the Admin API Key (format: `key_id:secret`)
+5. Set `GHOST_ADMIN_KEY=key_id:secret` in `.env`
+6. Set `GHOST_URL=https://your-ghost-site.com` in `.env`
+
+## How to Use
+
+Generate a weekly digest:
+
+```
+"Generate a weekly digest from my RSS feeds"
+"Create this week's newsletter"
+"Summarize my feeds from the last 7 days"
+```
+
+Choose a specific format:
+
+```
+"Create a topic deep dive about AI agents from this week's news"
+"Generate a curated picks digest for this week"
+"Write a weekly roundup newsletter"
+```
+
+Extend the date window:
+
+```
+"Generate a digest from the last 14 days"
+"Not many articles this week, extend the window to 2 weeks"
+```
+
+Publish to Ghost:
+
+```
+"Generate this week's digest and publish it to Ghost as a draft"
+"Create the newsletter and publish it to Ghost"
+```
+
+Output for Substack:
+
+```
+"Generate a newsletter digest for Substack"
+"Create the digest and give me the Markdown version"
+```
+
+## Output
+
+| Output | Description |
+|--------|-------------|
+| HTML | Ready to paste into any CMS |
+| Markdown | For Substack, Notion, Hashnode |
+| Plain text | For email clients |
+| Ghost draft | Published automatically if configured |
+
+## Substack Note
+
+Substack has no public API. The skill outputs a Markdown version of the digest for you to paste directly into the Substack editor.
+
+## Project Structure
+
+```
+newsletter-digest/
+├── SKILL.md
+├── README.md
+├── .env.example
+├── package.json
+├── feeds.json            (your RSS feed list, edit this)
+├── evals/
+│   └── evals.json
+├── references/
+│   ├── digest-format.md  (format rules, length targets, attribution)
+│   └── output-template.md (HTML templates for all 3 formats)
+└── scripts/
+    ├── fetch-feeds.js    (RSS fetching, dedup, date filtering)
+    └── ghost-publish.js  (Ghost Admin API posting)
+```
+
+## License
+
+MIT

package/skills/newsletter-digest/SKILL.md

@@ -0,0 +1,221 @@
+---
+name: newsletter-digest
+description: Aggregates RSS feeds from the past week, synthesizes the top stories using Gemini, and publishes a newsletter digest to Ghost CMS. Optionally outputs formatted Markdown for Substack or any other platform. Use when asked to generate a newsletter, create a weekly digest, summarize RSS feeds, compile top stories for a newsletter, or publish a digest to Ghost. Trigger when a user mentions newsletter digest, weekly roundup, RSS digest, compile top stories, or publish to Ghost.
+compatibility: [claude-code, gemini-cli, github-copilot]
+author: OpenDirectory
+version: 1.0.0
+---
+
+# Newsletter Digest
+
+Aggregate content from RSS feeds, synthesize the week's top stories with Gemini, and publish a digest to Ghost CMS. Supports weekly digests, topic-focused roundups, and curated picks formats.
+
+## Writing Style
+
+Apply to all generated digest content:
+
+- Active voice only
+- Short sentences, one idea per sentence
+- No em dashes — use a period or comma instead
+- No semicolons
+- No markdown or asterisks in the final newsletter body
+- No hashtags
+
+Banned words — do not use any of these in the digest:
+incredible, amazing, leveraging, synergize, game-changing, game changer, delve, harness, unlock, groundbreaking, cutting-edge, remarkable, paradigm, revolutionize, disruptive, transformative, thrilled, excited to share, powerful, innovative, comprehensive, actionable, crucial, vital, pivotal, elucidate, utilize, dive deep, tapestry, illuminate, revolutionize
+
+## CRITICAL RULE
+
+Every claim, statistic, and summary in the digest must come from the fetched articles. Never fabricate data points, quotes, or facts. If a source article has no meaningful content, skip it.
+
+---
+
+## Step 1: Check Setup
+
+**1a. Verify feeds are configured**
+
+Check if `feeds.json` exists in the skill directory:
+
+```bash
+ls /Users/ksd/Desktop/Varnan_skills/newsletter-digest/feeds.json 2>/dev/null
+```
+
+If missing, ask the user: "Which RSS feeds should I monitor? Share a list of URLs and I'll create feeds.json."
+
+The format is:
+```json
+[
+  { "url": "https://example.com/feed", "name": "Source Name" },
+  { "url": "https://another.com/rss", "name": "Another Source" }
+]
+```
+
+**1b. Verify required environment variables**
+
+Check that these are set:
+- `GEMINI_API_KEY` (required): for synthesis
+- `GHOST_URL` and `GHOST_ADMIN_KEY` (required for Ghost publishing, optional if output-only)
+
+If GEMINI_API_KEY is missing, stop and ask the user to set it.
+
+GHOST config is optional. If not set, the skill outputs formatted Markdown instead of publishing.
+
+**1c. Verify script dependencies**
+
+```bash
+ls /Users/ksd/Desktop/Varnan_skills/newsletter-digest/node_modules 2>/dev/null | head -1
+```
+
+If missing:
+
+```bash
+cd /Users/ksd/Desktop/Varnan_skills/newsletter-digest && npm install
+```
+
+---
+
+## Step 2: Fetch and Parse Feeds
+
+Run the fetch script:
+
+```bash
+node /Users/ksd/Desktop/Varnan_skills/newsletter-digest/scripts/fetch-feeds.js
+```
+
+The script:
+- Reads `feeds.json` for the list of feeds
+- Fetches each feed using `rss-parser`
+- Filters items published in the last 7 days
+- Deduplicates by URL
+- Sorts by publish date, newest first
+- Saves results to `/tmp/newsletter-digest-articles.json`
+
+If the script errors on a specific feed, it logs a warning and continues. One bad feed does not stop the run.
+
+After the script completes, verify the output:
+
+```bash
+node -e "
+const fs = require('fs');
+const data = JSON.parse(fs.readFileSync('/tmp/newsletter-digest-articles.json', 'utf8'));
+console.log('Total articles:', data.articles.length);
+console.log('From feeds:', data.feedsSummary.map(f => f.name + ': ' + f.count).join(', '));
+console.log('Date range:', data.dateRange.from, 'to', data.dateRange.to);
+"
+```
+
+If fewer than 3 articles are found, warn the user and ask if they want to extend the date range to 14 days. Pass `--days=14` to the script to widen the window.
+
+---
+
+## Step 3: Read Format Rules
+
+Read `references/digest-format.md` in full before generating any content. Internalize:
+- Three digest formats (Weekly Roundup, Topic Deep Dive, Curated Picks) and when to use each
+- Section structure and length requirements per format
+- Attribution rules (every claim links to a source)
+- Tone and voice guidelines
+
+Read `references/output-template.md` and select the template matching the requested format.
+
+If the user did not specify a format, default to Weekly Roundup.
+
+---
+
+## Step 4: Synthesize with Gemini
+
+Read `/tmp/newsletter-digest-articles.json` and build a synthesis prompt.
+
+**Prompt structure:**
+
+```
+You are a newsletter editor writing a digest of this week's top stories.
+
+Sources (use ONLY these — do not invent facts):
+[For each article: Title | Source | Date | Summary]
+
+Task:
+Write a [FORMAT] digest following this structure:
+[Paste the selected template from references/output-template.md]
+
+Rules:
+- Every claim, statistic, and quote must come from the sources above
+- Write in active voice, short sentences, contractions allowed
+- Do not use: incredible, amazing, leveraging, game-changing, delve, groundbreaking, cutting-edge, revolutionize
+- No em dashes, no semicolons, no hashtags
+- Include [Source: Name](URL) attribution after each major claim
+- Output plain text with HTML heading tags only (no markdown asterisks)
+```
+
+Call Gemini with this prompt. Target model: `gemini-2.0-flash` or the latest available.
+
+If the response contains any banned words or invented statistics not present in the source articles, regenerate with a stricter prompt.
+
+---
+
+## Step 5: Self-QA
+
+Run every check before presenting. Fix violations first.
+
+- [ ] Every claim, statistic, and quote traces to a source article
+- [ ] No banned words in any section
+- [ ] No em dashes in the text
+- [ ] No semicolons in the text
+- [ ] Every section has at least one source attribution with URL
+- [ ] Word count is within the format's target range (see references/digest-format.md)
+- [ ] No section repeated the same story twice
+- [ ] Title is specific (includes the date range or issue number, not just "Weekly Digest")
+- [ ] HTML output is valid (no unclosed tags if Ghost publishing)
+
+Fix all violations before presenting to the user.
+
+---
+
+## Step 6: Format and Preview
+
+Present the digest to the user in this order:
+
+1. **Title** — Issue date range or topic
+2. **Digest content** — Full formatted text
+3. **Source list** — All articles used, with URLs
+4. **Word count and article count**
+
+Ask: "Ready to publish to Ghost as a draft, or would you like to edit first?"
+
+---
+
+## Step 7: Publish to Ghost or Output
+
+**If GHOST_URL and GHOST_ADMIN_KEY are set and user confirms:**
+
+Run the publish script:
+
+```bash
+node /Users/ksd/Desktop/Varnan_skills/newsletter-digest/scripts/ghost-publish.js \
+  --title "DIGEST_TITLE" \
+  --status "draft"
+```
+
+The script reads `/tmp/newsletter-digest-output.json` (written by the agent in Step 6) and posts to Ghost.
+
+After posting, the script returns the Ghost admin URL for the draft. Present this to the user: "Draft saved at: [URL]. Review and publish from your Ghost dashboard."
+
+If the user wants to publish immediately, pass `--status "published"` to the script.
+
+**If Ghost is not configured or user says "output only":**
+
+Output the digest in three formats:
+
+1. **HTML** — ready to paste into any CMS
+2. **Markdown** — for Substack, Notion, or Hashnode
+3. **Plain text** — for email clients
+
+Add: "Substack does not offer a public API. Copy the Markdown version into your Substack editor directly."
+
+---
+
+## Notes on Substack
+
+Substack has no public API. Programmatic posting is not possible without reverse-engineering internal endpoints, which violates their terms. The skill outputs a formatted Markdown digest ready to paste into the Substack editor. This is the correct approach.
+
+

package/skills/newsletter-digest/evals/evals.json

@@ -0,0 +1,35 @@
+{
+  "skill_name": "newsletter-digest",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "Generate a weekly digest from my RSS feeds",
+      "expected_output": "Agent checks feeds.json exists and GEMINI_API_KEY is set. Runs scripts/fetch-feeds.js. Verifies output JSON contains at least 3 articles. Reads references/digest-format.md and references/output-template.md. Calls Gemini with the fetched articles and Weekly Roundup template. Returns a digest with a Top Story section (100-150 words), 3+ Also Worth Reading entries (40-60 words each), Quick Takes bullets, and a closing line. Every claim in the digest has a source attribution with a real URL from the fetched articles. No banned words. No invented statistics. Word count is 350-500 words. Asks user to confirm before publishing to Ghost.",
+      "files": []
+    },
+    {
+      "id": 2,
+      "prompt": "Create a topic deep dive newsletter about AI agent frameworks from this week's news",
+      "expected_output": "Agent runs fetch-feeds.js. Filters fetched articles to those related to AI agents and frameworks. Reads digest-format.md and selects Topic Deep Dive format. Calls Gemini with filtered articles and Topic Deep Dive template. Returns a digest with 'The Big Picture' section, 3-4 Key Development sections each with a linked article title, a 'What to Watch' bullet list, and a sources list. All claims trace to fetched articles. If fewer than 3 relevant articles are found, agent warns the user and asks if they want to extend the window to 14 days (re-run with --days=14).",
+      "files": []
+    },
+    {
+      "id": 3,
+      "prompt": "Generate this week's digest and publish it to Ghost as a draft",
+      "expected_output": "Agent checks GHOST_URL and GHOST_ADMIN_KEY are set. Runs fetch-feeds.js. Synthesizes digest. Writes digest HTML and metadata to /tmp/newsletter-digest-output.json. Runs scripts/ghost-publish.js with --title and --status=draft. Script generates Ghost JWT from GHOST_ADMIN_KEY (id:secret format, hex-decoded, HS256, 5-minute expiry). POSTs to /ghost/api/admin/posts/?source=html. Script returns the Ghost admin editor URL. Agent presents the URL to the user: 'Draft saved at [URL]. Review and publish from your Ghost dashboard.'",
+      "files": []
+    },
+    {
+      "id": 4,
+      "prompt": "Generate a newsletter digest for Substack",
+      "expected_output": "Agent generates the digest normally. When presenting output, provides three formats: HTML (for any CMS), Markdown (for Substack), and plain text. Includes a clear note: 'Substack does not offer a public API. Copy the Markdown version into your Substack editor directly.' Does not attempt any Substack API calls. The Markdown version uses ## headings, [text](url) links, and - bullet points.",
+      "files": []
+    },
+    {
+      "id": 5,
+      "prompt": "The feeds only returned 2 articles this week — can you extend the window?",
+      "expected_output": "Agent re-runs fetch-feeds.js with --days=14 flag. Fetches articles from the last 14 days instead of 7. Reports the new article count. Continues with synthesis using the extended article set. If still fewer than 3 articles after 14 days, agent warns that the feeds may not be updating frequently and asks the user to check that the feed URLs are correct.",
+      "files": []
+    }
+  ]
+}

package/skills/newsletter-digest/feeds.json

@@ -0,0 +1,7 @@
+[
+  { "url": "https://feeds.feedburner.com/TheHackersNews", "name": "The Hacker News" },
+  { "url": "https://hnrss.org/frontpage", "name": "Hacker News Front Page" },
+  { "url": "https://www.infoq.com/feed/", "name": "InfoQ" },
+  { "url": "https://changelog.com/feed", "name": "Changelog" },
+  { "url": "https://css-tricks.com/feed/", "name": "CSS-Tricks" }
+]

package/skills/newsletter-digest/package.json

@@ -0,0 +1,15 @@
+{
+  "name": "newsletter-digest",
+  "version": "1.0.0",
+  "description": "RSS aggregation and newsletter digest generation scripts",
+  "type": "module",
+  "scripts": {
+    "fetch": "node scripts/fetch-feeds.js",
+    "fetch:extended": "node scripts/fetch-feeds.js --days=14",
+    "publish": "node scripts/ghost-publish.js"
+  },
+  "dependencies": {
+    "jsonwebtoken": "^9.0.2",
+    "rss-parser": "^3.13.0"
+  }
+}