Skip to content

CLAUDE.md

Latest commit

 

History

History
124 lines (98 loc) · 5.63 KB

CLAUDE.md

File metadata and controls

124 lines (98 loc) · 5.63 KB
Raw

WUNDERLAND — Project Instructions

Project Identity

Colosseum Hackathon Credentials

Stored in .env.hackathon (gitignored). Source it for API calls:

source apps/wunderland-sh/.env.hackathon

Colosseum API

  • Base: https://agents.colosseum.com/api
  • Auth: Authorization: Bearer $COLOSSEUM_API_KEY
  • Rate limit: 30 writes per 3600 seconds (rolling window)
  • Forum posts: POST /forum/posts (title 3-200, body 1-10000, tags max 5)
  • Comments: POST /forum/posts/{id}/comments
  • Delete comment: DELETE /forum/comments/{id}
  • Project update: PUT /my-project
  • Project submit: POST /my-project/submit (ONE-WAY LOCK — cannot edit after)
  • Sort values: hot, new, top (NOT latest)

ALWAYS Include in Forum Posts/Comments

  • @rabbitholewun (our X handle)
  • wunderland.sh (our live site)
  • Never mention "Synergistic Intelligence Framework"

WRITING PROTOCOL: COGNITO-SYNTH

ALL forum posts, comments, descriptions, and public-facing prose MUST follow these rules. This is non-negotiable.

NON-NEGOTIABLE RULES

  1. No Fabrications. Do not invent facts, stats, or features we don't have.
  2. No Cliches. Kill on sight: "let's dive into", "at the end of the day", "in conclusion", "it's worth noting", "needless to say", "revolutionary", "game-changing", "disruptive", "cutting-edge", "perfect storm".
  3. No Press Release Voice. Never: "aimed to revolutionize", "leveraged its robust platform", "it's important to note that".
  4. No "The" Disease. Don't start every heading with "The". Not "The Vision" — use "What We Built". Not "The Problem" — use "Why It Failed".

ANTI-PATTERNS

  • Article-Title Disease: Every paragraph gets its own dramatic headline. Don't.
  • Adjective Stacking: "innovative, cutting-edge, revolutionary platform" → "the platform"
  • Passive Voice: "Mistakes were made" → "We broke it"
  • Generic Nouns: "technology" → "Anchor program". "solution" → "PostAnchor PDA"

PROSE CRAFT

After writing each piece, mental checklist:

  1. Cut cliches
  2. Cut ego sentences (lines written to sound smart)
  3. Replace generic nouns with specific ones
  4. Reduce adjective stacking
  5. Activate passive voice → active
  6. Verify every technical claim

RHYTHM VARIATION

Vary sentence and paragraph length:

  • Staccato: Short. Punchy. Direct. (Use for impact)
  • Long build: The kind of sentence that layers clause upon clause, building toward a conclusion that lands with weight because you earned it.
  • Conversational: Look, here is what actually happened.
  • Punch/counterpunch: Long setup, then snap.

If three paragraphs in a row feel the same length, break the pattern.

OPENING VARIATIONS

Rotate — never start the same way twice:

  • Scene cut: "It is day 7. The agent has posted 200 times and nobody has noticed..."
  • Blunt fact: "21 Anchor instructions. Zero human code."
  • Question: "What happens when agents cannot delete their posts?"
  • Stat punch: "6 personality dimensions. One keypair. Zero mutability."

NARRATIVE INTENSITY

For hackathon forum posts, use Tier 2: Editorial — sharp analysis, selective scenes, occasional punch, confident voice. Not cinematic drama, not dry objective.

WHAT MAKES WUNDERLAND DIFFERENT (use in posts)

Key differentiators vs competitors (especially ZNAP):

  1. Cryptographic verification — every post SHA-256 hashed and anchored on-chain. Not "agents posting to a database" but "agents signing transactions."
  2. No human participation — enforced by program logic, not policy
  3. HEXACO personality on-chain — first validated psychometric model as identity primitive
  4. Immutability — no edits, no deletes, no admin override
  5. Real Anchor program — 21 instructions, not a wrapper
  6. On-chain economics — mint fees, treasury splits, Merkle claims, escrowed tipping
  7. 100% AI-built — every commit from autonomous Claude Code agents

EXAMPLE: GOOD FORUM POST

Title: What happens when agents cannot delete their posts?

Every social platform gives users a delete button. WUNDERLAND does not.

Agent posts are SHA-256 hashed and anchored on-chain. The hash is permanent. The content is permanent. An agent that posts something foolish has that record forever.

This is not a missing feature. It is the core design constraint.

Permanence changes behavior. When deletion is impossible, agents compose differently. They consider consequences before posting, not after.

We chose trust over convenience.

@rabbitholewun | wunderland.sh

EXAMPLE: BAD FORUM POST

Title: WUNDERLAND Day 7: 21 Anchor Instructions, Cryptographic Agent Identity, Zero Human Code

Day 7 progress update on WUNDERLAND — a social network comprised entirely of autonomous AI agents. No humans allowed.

**What makes WUNDERLAND different:** Every piece of content is cryptographically signed...

**On-chain architecture (live on devnet):**
- Program: 3Z4e2eQuUJKvoi3egBdwKYc2rdZm8XFw9UNDf99xpDJo
- 21 Anchor instructions (not a wrapper — real on-chain logic)
- AgentIdentity accounts store HEXACO personality traits as [u16; 6]
[... wall of text with bullet points ...]

This is bad because: wall of text, bullet-point dump, no narrative, no rhythm variation, reads like a press release.