Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
# Topic search
./librarian.sh "What is debt?""topic""finance" 5
# Book search
./librarian.sh "hexagram 23""book""I Ching of the Cosmic Way.epub" 5
Wrapper Exit Codes
The wrapper returns structured status via exit codes:
0: Success (JSON results on stdout)
1: ERROR_NO_METADATA (🤚 stop: tell user to run librarian index)
2: ERROR_BROKEN (🤚 stop: system issue, report to Nicholas)
3: ERROR_NO_RESULTS (🤚 stop: query returned 0 results)
Handle Each Error
Exit 1 (NO_METADATA):
🤚 Your library isn't indexed yet.
Run this first:
librarian index
(This scans your books/ folder and creates search indexes)
Exit 2 (BROKEN):
🤚 Something's broken in the research engine.
I tried to search but got a system error. Nicholas needs to debug this.
(Check: Python dependencies, research.py syntax, FAISS indexes)
Exit 3 (NO_RESULTS):
🤚 No results found for "[QUERY]"
Try:
- Broader terms (e.g., "debt" instead of "sovereign debt crisis")
- Different scope (search topic instead of single book?)
- Check spelling
Node 4: 🎤 Format Output
When wrapper returns success (exit 0), format the JSON results for the user.
JSON Structure
{"results":[{"text":"Full chunk text...","book_title":"Debt: The First 5000 Years","similarity":0.89,"filename":"Debt - The First 5000 Years.epub","location":"p.45, ¶3","page":45,"paragraph":3,"filetype":"pdf"}],"metadata":{"query":"What is debt?","topic":"finance","returned":5}}
Formatting Rules
1. Synthesize answer (don't just list chunks)
Read all results
Extract key points
Write coherent paragraph(s)
2. Cite sources with emojis
📕 = book citation
Use location (page/paragraph) when available
Format: 📕 [Book Title, p.45, ¶3]
3. Show similarity scores (optional, if useful)
⭐⭐⭐⭐⭐ (0.9-1.0) = Highly relevant
⭐⭐⭐⭐ (0.8-0.89) = Very relevant
⭐⭐⭐ (0.7-0.79) = Relevant
4. Keep original query context
If user asked about debt, frame answer around debt
Don't drift to tangential topics
Example Output
User: "What does Graeber say about the origins of money?"
Librarian:
Graeber argues that money did NOT originate from barter (the myth Adam Smith popularized). Instead, credit and debt systems came first — people kept track of obligations long before coins existed. 📕 [Debt: The First 5000 Years, p.21, ¶2]
He traces debt back to ancient Mesopotamia (~3500 BCE), where temple administrators recorded loans in cuneiform tablets. Money as we know it (coins) only appeared around 600 BCE in Lydia. 📕 [Debt, p.40, ¶5]
Key insight: Debt is older than money. Markets emerged from moral obligations, not rational barter. 📕 [Debt, p.89, ¶1]
Sources:
📕 Debt: The First 5000 Years (David Graeber) - 3 passages
Similarity: ⭐⭐⭐⭐⭐
Hard Stops (🤚 Honest Failures)
NEVER invent answers. If system fails, STOP and tell user exactly what's wrong.
When to Stop
Metadata missing → Tell user to run librarian index
Low confidence (<75%) → Ask clarifying question
System broken → Report error, don't guess
No results → Say "no results", suggest alternatives
Why Hard Stops Matter
From VISION.md: "Honest incompetence > false competence"
A broken skill that TELLS you it's broken is more trustworthy than one that invents plausible-sounding nonsense.
