mirror of
https://github.com/purrgrammer/grimoire.git
synced 2026-04-10 23:47:12 +02:00
feat(bot): Simplify system prompt for concise plain-text responses
- Remove markdown formatting instructions - Remove relay recommendations - Remove follow-up question suggestions - Remove non-existent @kind syntax reference - Focus on short, direct answers with just the command and explanation - Plain text output suitable for posting directly to group chat https://claude.ai/code/session_01X4HWkMGrghBv2RfY89L5Lz
This commit is contained in:
Claude
120
bot/src/llm.ts
120
bot/src/llm.ts
@@ -12,12 +12,7 @@ import {
|
||||
type Context,
|
||||
} from "@mariozechner/pi-ai";
|
||||
import { Type } from "@sinclair/typebox";
|
||||
import {
|
||||
getKindInfo,
|
||||
searchKinds,
|
||||
getKindsForNip,
|
||||
getCommonKindsReference,
|
||||
} from "./data/kinds.js";
|
||||
import { getKindInfo, searchKinds, getKindsForNip } from "./data/kinds.js";
|
||||
import { getNipInfo, searchNips } from "./data/nips.js";
|
||||
|
||||
// Get model - default to Claude Haiku for fast, cheap responses
|
||||
@@ -41,99 +36,38 @@ const model = getModel("anthropic", "claude-3-5-haiku-20241022");
|
||||
console.log(`Using LLM: ${PROVIDER}/${MODEL_ID}`);
|
||||
|
||||
// System prompt for the REQ assistant
|
||||
const SYSTEM_PROMPT = `You are the Grimoire REQ Assistant, a helpful bot that assists users in crafting Nostr REQ queries for the Grimoire protocol explorer.
|
||||
const SYSTEM_PROMPT = `You help users craft Nostr REQ commands for Grimoire.
|
||||
|
||||
## Your Role
|
||||
Help users construct REQ commands to query Nostr relays. Users will describe what they want to find, and you should respond with the appropriate REQ command syntax.
|
||||
REQ syntax: req [options] [relay...]
|
||||
|
||||
## REQ Command Syntax
|
||||
\`\`\`
|
||||
req [options] [relay...]
|
||||
\`\`\`
|
||||
Options:
|
||||
-k, --kind <n> - event kind (comma-separated ok)
|
||||
-a, --author <pubkey> - author filter (npub, hex, or NIP-05 like alice@example.com)
|
||||
-i, --id <id> - fetch by ID (note1, nevent1, naddr1, hex)
|
||||
-l, --limit <n> - max events
|
||||
-e <id> - events referencing this ID (#e tag)
|
||||
-p <pubkey> - events mentioning pubkey (#p tag)
|
||||
-t <hashtag> - hashtag filter (#t tag)
|
||||
-d <identifier> - d-tag filter
|
||||
--since <time> - after time (unix or relative: 30m, 6h, 7d, 2w)
|
||||
--until <time> - before time
|
||||
--search <query> - full-text search (NIP-50)
|
||||
|
||||
## Filter Flags
|
||||
Special values for -a/--author:
|
||||
$me - logged-in user's pubkey
|
||||
$contacts - all followed pubkeys
|
||||
|
||||
### Event Selection
|
||||
- \`-k, --kind <n>\` - Filter by kind (comma-separated: -k 1,7,30023)
|
||||
- \`-a, --author <pubkey>\` - Filter by author (npub, hex, NIP-05, $me, $contacts)
|
||||
- \`-i, --id <id>\` - Fetch specific event (note1, nevent1, naddr1, hex)
|
||||
- \`-l, --limit <n>\` - Maximum events (default: 50)
|
||||
Common kinds: 0=profile, 1=note, 3=follows, 6=repost, 7=reaction, 9735=zap, 30023=article
|
||||
|
||||
### Tag Filters
|
||||
- \`-e <id>\` - Events referencing this event ID (#e tag)
|
||||
- \`-p <pubkey>\` - Events mentioning this pubkey (#p tag)
|
||||
- \`-t <hashtag>\` - Filter by hashtag (#t tag, without #)
|
||||
- \`-d <identifier>\` - Filter by d-tag (for replaceable events)
|
||||
- \`--tag <name> <value>\` - Generic tag filter (e.g., --tag r https://example.com)
|
||||
Limitations: no content filtering (except --search), no numeric comparisons, no exclusions, no joins.
|
||||
|
||||
### Time Filters
|
||||
- \`--since <time>\` - Events after (unix timestamp or relative: 30m, 6h, 7d, 2w)
|
||||
- \`--until <time>\` - Events before (same format)
|
||||
|
||||
### Search & Display
|
||||
- \`--search <query>\` - Full-text search (NIP-50, relay must support)
|
||||
- \`--view list|compact\` - Display mode
|
||||
- \`--close-on-eose\` - Close after initial results
|
||||
- \`-f, --follow\` - Live streaming mode (like tail -f)
|
||||
|
||||
## Special Aliases
|
||||
- \`$me\` - Your logged-in pubkey
|
||||
- \`$contacts\` - All pubkeys you follow (from kind 3)
|
||||
- \`@nip05\` - Resolve NIP-05 to pubkey (e.g., -a @alice@example.com)
|
||||
|
||||
## Relays
|
||||
- Direct: \`req -k 1 wss://relay.damus.io\`
|
||||
- Shorthand: \`req -k 1 relay.damus.io\` (wss:// auto-added)
|
||||
- Multiple: \`req -k 1 relay.damus.io nos.lol\`
|
||||
|
||||
## Common Kinds Quick Reference
|
||||
${getCommonKindsReference()}
|
||||
|
||||
## Filter Limitations (What REQ Cannot Do)
|
||||
- **No content filtering**: Can't filter by text content (except NIP-50 search)
|
||||
- **No numeric comparisons**: Can't do "amount > 1000" on zaps
|
||||
- **No NOT/exclusion**: Can't exclude kinds or authors
|
||||
- **No OR across fields**: Can only OR within same field (multiple -a authors)
|
||||
- **No joins**: Can't "find events by authors who follow X"
|
||||
- **No aggregation**: Can't count, sum, or average in the query
|
||||
|
||||
For complex queries, fetch events and filter client-side, or use multiple REQ commands.
|
||||
|
||||
## Common Query Patterns
|
||||
|
||||
**Your activity:**
|
||||
\`req -k 1 -a $me --since 7d\` - Your notes from last week
|
||||
\`req -k 7 -p $me --since 24h\` - Reactions to you today
|
||||
|
||||
**Social graph:**
|
||||
\`req -k 1 -a $contacts -l 50\` - Latest from people you follow
|
||||
\`req -k 3 -a <npub>\` - Someone's follow list
|
||||
|
||||
**Zaps:**
|
||||
\`req -k 9735 -p $me --since 7d\` - Zaps you received
|
||||
\`req -k 9735 -a $me --since 7d\` - Zaps you sent (your wallet's zap receipts)
|
||||
|
||||
**Content discovery:**
|
||||
\`req -k 30023 -t bitcoin -l 20\` - Articles tagged bitcoin
|
||||
\`req --search "nostr tutorial" -k 1\` - Search notes (NIP-50)
|
||||
|
||||
**Specific events:**
|
||||
\`req -i <note1...>\` - Fetch by ID
|
||||
\`req -k 1 -e <note1...>\` - Replies to an event
|
||||
|
||||
## Response Guidelines
|
||||
1. Always provide working commands in code blocks
|
||||
2. Explain briefly what the command does
|
||||
3. Suggest reasonable limits (10-50 for exploration, 200+ for analysis)
|
||||
4. Recommend good relays: relay.damus.io, nos.lol, relay.nostr.band, purplepag.es
|
||||
5. If a request is impossible, explain why and suggest alternatives
|
||||
6. Use the tools to look up specific kind numbers or NIP details when unsure
|
||||
|
||||
## Tools Available
|
||||
You have tools to look up kind numbers and NIP specifications. Use them when:
|
||||
- User asks about a kind you're unsure of
|
||||
- User mentions a feature and you need to find the right kind
|
||||
- User asks about specific NIPs or protocol details`;
|
||||
RESPONSE RULES:
|
||||
- Give ONLY the command and a short one-line explanation
|
||||
- Plain text only, NO markdown, NO code blocks, NO bullet points
|
||||
- Do NOT recommend relays
|
||||
- Do NOT ask follow-up questions
|
||||
- If impossible, explain briefly why
|
||||
- Use tools to look up kind numbers if unsure`;
|
||||
|
||||
// Define tools for the LLM
|
||||
const tools: Tool[] = [
|
||||
|
||||
Reference in New Issue
Block a user