sawz/last30days-skill
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refocus skill on prompt-first output with stats summary

Browse Source 
- Parse user intent for TOPIC and TARGET_TOOL upfront
- Ask follow-up if target tool unclear
- Show impressive stats summary after research (upvotes, likes, etc.)
- Primary output is now copy-paste-ready prompts for target tool
- Keep expert context for follow-up custom prompt requests
- Research is internalized, not dumped back at user

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Matt Van Horn
2026-01-23 13:26:01 -08:00
parent 5ca4829be4
commit c037eea15f
1 changed files with 129 additions and 72 deletions
Download Patch File Download Diff File Expand all files Collapse all files
SKILL.md
+141 -84
View File
@@ -1,20 +1,50 @@
---
name: last30days
description: Research a topic from the last 30 days on Reddit + X; judge/summarize into best practices, a prompt pack, and a reusable context snippet.
argument-hint: "[topic]"
description: Research a topic from the last 30 days on Reddit + X, become an expert, and write copy-paste-ready prompts for the user's target tool.
argument-hint: "[topic] for [tool]" or "[topic]"
context: fork
agent: Explore
disable-model-invocation: true
allowed-tools: Bash, Read, Write
allowed-tools: Bash, Read, Write, AskUserQuestion
---
# last30days: 30-Day Research Synthesis
# last30days: Become Expert → Write Prompts
Research a topic across Reddit and X from the last 30 days, then synthesize findings into actionable best practices, prompts, and reusable context.
Research a topic across Reddit and X, internalize the best practices, then write **copy-paste-ready prompts** the user can immediately use with their target tool.
## CRITICAL: Parse User Intent
Before doing anything, parse the user's input for TWO things:
1. **TOPIC**: What they want to learn about (e.g., "web app mockups", "Claude Code skills", "image generation")
2. **TARGET TOOL**: Where they'll use the prompts (e.g., "Nano Banana Pro", "ChatGPT", "Claude", "Midjourney")
Common patterns:
- `[topic] for [tool]` → "web mockups for Nano Banana Pro"
- `[topic] prompts for [tool]` → "UI design prompts for Midjourney"
- `[tool] [topic]` → "Nano Banana Pro dashboard mockups"
- Just `[topic]` → Ask follow-up
**If TARGET TOOL is unclear**, use AskUserQuestion:
```
What tool will you use these prompts with?
Options:
1. Nano Banana Pro (image generation)
2. ChatGPT / Claude (text/code)
3. Midjourney / DALL-E (image generation)
4. Other (tell me)
```
**Store these values mentally** - you'll need them for the entire conversation:
- `TOPIC = [extracted topic]`
- `TARGET_TOOL = [extracted tool]`
---
## Setup Check
First, verify API key configuration exists:
Verify API key configuration exists:
```bash
if [ ! -f ~/.config/last30days/.env ]; then
@@ -26,11 +56,10 @@ fi
### If SETUP_NEEDED
Run the NUX flow to configure API keys. Use the AskUserQuestion tool to collect:
Run NUX flow to configure API keys. Use AskUserQuestion to collect:
1. **OpenAI API Key** (optional but recommended for Reddit research)
2. **xAI API Key** (optional but recommended for X research)
3. **Model policies** (optional, defaults are usually fine)
Then create the config:
@@ -42,12 +71,6 @@ cat > ~/.config/last30days/.env << 'ENVEOF'
OPENAI_API_KEY=
XAI_API_KEY=
# Model selection (optional)
# OPENAI_MODEL_POLICY=auto|pinned (default: auto)
# OPENAI_MODEL_PIN=gpt-5.2 (only if pinned)
# XAI_MODEL_POLICY=latest|stable|pinned (default: latest)
# XAI_MODEL_PIN=grok-4 (only if pinned)
ENVEOF
chmod 600 ~/.config/last30days/.env
@@ -55,91 +78,125 @@ echo "Config created at ~/.config/last30days/.env"
echo "Please edit it to add your API keys, then run the skill again."
```
After creating the file, instruct the user to edit `~/.config/last30days/.env` and add their keys.
**STOP HERE if setup was needed. Do not proceed until keys are configured.**
**STOP HERE if setup was needed.**
---
## Research Execution
If configured, run the research orchestrator:
Run the research orchestrator with the TOPIC:
```bash
python3 ~/.claude/skills/last30days/scripts/last30days.py "$ARGUMENTS" --emit=compact 2>&1
```
The script will:
- Auto-detect which keys are available
- Auto-select the best models (or use pinned versions)
- Search Reddit via OpenAI Responses API (if OpenAI key present)
- Search X via xAI Responses API (if xAI key present)
- Enrich Reddit threads with real engagement metrics
- Score, rank, and dedupe results
- Output files to `~/.local/share/last30days/out/`
---
## RESEARCH DATA
## FIRST: Show the Work (Stats Summary)
The output above contains the research data. Now synthesize it.
**Before anything else**, aggregate the metrics from the research and display an impressive summary. Parse the output above and calculate:
---
- Count of Reddit threads
- Sum of all Reddit upvotes (pts)
- Sum of all Reddit comments (cmt)
- Count of X posts
- Sum of all X likes
- Sum of all X reposts (rt)
- List unique subreddits
- List unique X authors
## Your Role: Judge and Synthesizer
Display it in this format:
You are now the expert judge. Using the research data above, produce:
### A) Best Practices (Grouped & Actionable)
Group findings into 3-7 thematic categories. For each best practice:
- State the practice clearly and actionably
- Cite supporting item IDs (e.g., "supported by R3, R7, X2")
- Note if it's **strongly supported** (multiple high-score sources) or **niche** (single source or low engagement)
### B) Prompt Pack (3-7 Copy/Paste Prompts)
Create ready-to-use prompts tailored to the topic. Each prompt should:
- Be immediately usable (copy/paste ready)
- Target a specific use case discovered in the research
- Include any relevant context or constraints from the findings
### C) Reusable Context Snippet
Create a compact (~200-400 words) context block that other skills/tools can import. Include:
- Core concepts and terminology
- Key techniques or patterns
- Common pitfalls to avoid
- Brief source attribution
### D) Sources Appendix
List all source URLs organized by platform:
- **Reddit**: Title, subreddit, URL, score
- **X**: Author, text excerpt, URL, engagement
### E) Confidence Assessment
Explicitly state:
- **Strongly Supported**: Practices backed by multiple high-engagement sources
- **Emerging/Niche**: Practices from single sources or low engagement (still valuable but use with awareness)
- **Gaps**: What the research didn't cover well
---
## Final Output
After completing your synthesis:
1. Display the full report to the user
2. Confirm the files were written:
- `~/.local/share/last30days/out/report.md`
- `~/.local/share/last30days/out/report.json`
- `~/.local/share/last30days/out/last30days.context.md`
3. Show the header summary:
```
Models used: OpenAI={model} xAI={model}
Mode: {reddit-only|x-only|both}
Coverage: {note about triangulation if single-source}
📊 Research Complete
Analyzed {total_sources} sources from the last 30 days
├─ Reddit: {n} threads │ {sum} upvotes │ {sum} comments
├─ X: {n} posts │ {sum} likes │ {sum} reposts
└─ Top voices: r/{sub1}, r/{sub2}, @{handle1}, @{handle2}
Now synthesizing into expert knowledge...
```
**Use real numbers from the research output.** This shows the user the skill actually did work.
---
## THEN: Internalize the Research
Read the research output above. You are now becoming an **expert** in this topic.
Your job is NOT to dump the research back at the user. Your job is to:
1. **Absorb** all the patterns, techniques, and insights
2. **Synthesize** them into expertise
3. **Apply** that expertise to write prompts for the user's TARGET_TOOL
---
## PRIMARY OUTPUT: Copy-Paste Prompts for TARGET_TOOL
**This is the main deliverable.** Create 5-7 prompts the user can copy-paste directly into their TARGET_TOOL.
### Format Each Prompt:
```
### [Use Case Name]
**When to use:** [1-line description]
**Prompt:**
```
[The actual prompt they copy-paste - ready to use, no placeholders unless clearly marked with [brackets]]
```
**Why this works:** [1-line explaining what research insight this is based on, cite source ID]
```
### Prompt Quality Checklist:
- [ ] Can be pasted directly into TARGET_TOOL with zero edits
- [ ] Uses specific patterns/keywords discovered in research
- [ ] Appropriate length and style for TARGET_TOOL
- [ ] Covers the most common use cases for TOPIC
---
## SECONDARY: Brief Best Practices (Optional)
Only include if the user seems to want background. Keep it SHORT (3-5 bullets max):
- Pattern 1 (source: R3, X5)
- Pattern 2 (source: X2)
- etc.
---
## FOLLOW-UP OFFER
After delivering prompts, ALWAYS ask:
> **Want me to write a custom prompt?** Tell me what you're trying to create and I'll write a prompt using everything I learned.
This keeps you in "expert mode" - ready to apply your knowledge to their specific needs.
---
## CONTEXT MEMORY
For the rest of this conversation, remember:
- **TOPIC**: {topic}
- **TARGET_TOOL**: {tool}
- **KEY PATTERNS**: {list the top 3-5 patterns you learned}
When the user asks for another prompt later, you don't need to re-research. Apply what you learned.
---
## Output Summary Footer
End with a compact reminder of what you learned:
```
---
📚 Expert in: {TOPIC} for {TARGET_TOOL}
📊 Based on: {n} Reddit threads ({sum} upvotes) + {n} X posts ({sum} likes)
🎯 Ready for custom prompts - just tell me what you want to create.
```