BankrBot · cryptomfer · Jun 17, 2026
diff --git a/cc0company/agent-services/SKILL.md b/cc0company/agent-services/SKILL.md
@@ -0,0 +1,231 @@
+---
+name: cc0company-agent-services
+version: 1.0.0
+description: Pay-per-call AI image generation on cc0.company via x402 v2 USDC. Five fine-tuned CC0 LoRA models — sartoshi-gen, darkfarms-gen, hokusai-gen, van-gogh-gen, monet-gen — outputs released into the public domain.
+homepage: https://cc0.company
+api_base: https://cc0.company/api
+chain: base
+chain_id: 8453
+---
+
+# cc0.company Agent Services — AI Image Generation
+
+Five fine-tuned CC0 image-generation models, accessible via x402 v2
+on Base mainnet. Agents pay **0.069 USDC per image**, receive a
+job_id immediately, poll until the generation is ready, and use the
+output anywhere — every output is CC0 (no licensing, no attribution
+required).
+
+## The 5 models
+
+| Slug | Style | Trained by | Per-model prompt guide |
+|---|---|---|---|
+| `sartoshi-gen` | Hand-drawn comics — mfer stick figures, pepe frogs, NFT in-jokes | cc0toshi | [`./sartoshi-gen.md`](./sartoshi-gen.md) |
+| `darkfarms-gen` | Crypto pepe meme art (Darkfarms1 style) | cc0toshi | [`./darkfarms-gen.md`](./darkfarms-gen.md) |
+| `hokusai-gen` | Edo-period ukiyo-e woodblock prints | cc0toshi | [`./hokusai-gen.md`](./hokusai-gen.md) |
+| `van-gogh-gen` | Post-impressionist Van Gogh paintings | cc0toshi | [`./van-gogh-gen.md`](./van-gogh-gen.md) |
+| `monet-gen` | French impressionist Monet paintings | cc0toshi | [`./monet-gen.md`](./monet-gen.md) |
+
+**Critical:** before invoking a model, fetch its per-model prompt
+guide. Every LoRA was trained on a specific caption register and the
+guide documents the canonical templates. Skipping it typically costs
+2-3× in output quality.
+
+## How it works (3 calls)
+
+```bash
+# 1. Browse the catalogue
+curl https://cc0.company/api/store/agent-services
+
+# 2. Invoke a model (x402-gated, ~0.069 USDC)
+curl -X POST https://cc0.company/api/store/agent-services/sartoshi-gen/invoke \
+  -H "Content-Type: application/json" \
+  -H "PAYMENT-SIGNATURE: <base64-signed-payload>" \
+  -H "X-Agent-Name: my_agent" \
+  -d '{"prompt": "<your-prompt>"}'
+# → 202 { "job_id": "agentservicejob_xxx", "agent": { "name", "api_key", "was_new" } }
+
+# 3. Poll until done
+curl https://cc0.company/api/store/agent-services/jobs/agentservicejob_xxx
+# → { "job": { "status": "succeeded", "output_url": "https://..." } }
+```
+
+## Endpoints
+
+| Method | Path | Auth | Purpose |
+|---|---|---|---|
+| `GET` | `/api/store/agent-services` | none | List all 5 models + `prompt_guide_url` per model |
+| `GET` | `/api/store/agent-services/{slug}` | none | One model's full detail |
+| `POST` | `/api/store/agent-services/{slug}/invoke` | x402 v2 | Pay → start a generation. Returns 202 + `job_id` |
+| `GET` | `/api/store/agent-services/jobs/{jobId}` | none | Poll a job. Status: `processing` / `succeeded` / `failed` / `refunded`. **IPFS pinning is automatic** — every succeeded job ships with `ipfs_persisted: true` + `ipfs_url` baked into the response. |
+| `POST` | `/api/store/agent-services/{slug}/pay-and-invoke` | tx_hash | Human path: send USDC, pass tx_hash, no x402 lib needed |
+
+> **`/jobs/:jobId/persist` is deprecated.** Earlier versions of the
+> platform required a separate 0.01 USDC pay-to-pin call to move the
+> output from the ephemeral Replicate URL (~1h TTL) to a permanent
+> IPFS pin. That step is now automatic — every succeeded generation
+> is pinned to IPFS before the job transitions to `succeeded`. The
+> `/persist` route remains as a no-op for backward compatibility but
+> charges nothing and does nothing. Don't call it on new
+> integrations.
+
+## Pricing
+
+- **Invoke:** 0.069 USDC per image, fixed across all 5 models. The
+  price comes from the live `paymentRequired.maxAmountRequired`
+  field on the 402 challenge — never hard-code it; read it at
+  runtime.
+- **IPFS pinning:** **free + automatic**. Every succeeded job is
+  pinned to IPFS as part of the polling flow. The `output_url` you
+  get back IS the IPFS gateway URL; `ipfs_url` carries the canonical
+  `ipfs://` form. No extra payment, no extra call.
+
+## x402 v2 client setup
+
+The canonical pattern uses the `@x402/fetch` + `@x402/evm` packages
+with a viem signer. Three lines of setup, then any fetch through
+the wrapper handles the 402 → sign → retry cycle automatically.
+
+```typescript
+import { x402Client, wrapFetchWithPayment } from "@x402/fetch"
+import { registerExactEvmScheme } from "@x402/evm/exact/client"
+import { privateKeyToAccount } from "viem/accounts"
+
+// Any viem-compatible signer works:
+// - throwaway: const pk = generatePrivateKey()
+// - CDP wallet export
+// - imported pk from env
+const signer = privateKeyToAccount(YOUR_PK as `0x${string}`)
+const client = new x402Client()
+registerExactEvmScheme(client, { signer })
+const fetchWithPayment = wrapFetchWithPayment(fetch, client)
+
+// Use exactly like fetch — the wrapper auto-pays 402 challenges
+const res = await fetchWithPayment(
+  "https://cc0.company/api/store/agent-services/sartoshi-gen/invoke",
+  {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "X-Agent-Name": "my_agent",  // optional, see "auto-register" below
+    },
+    body: JSON.stringify({ prompt: "..." }),
+  }
+)
+const job = await res.json()
+```
+
+If your runtime can't run Node packages (e.g. Bankr, openclaw), see
+[`../x402-payments/SKILL.md`](../x402-payments/SKILL.md) for the
+HTTP-only signing pattern.
+
+## Auto-registration (first paid invoke)
+
+If the wallet you sign with has never paid cc0.company before AND
+you don't send an `X-Agent-API-Key` header, the 202 response
+carries:
+
+```json
+{
+  "agent": {
+    "name": "my_agent",          // or "agent_<8-hex-wallet-prefix>" by default
+    "api_key": "cc0_agent_...",   // shown ONCE — persist it
+    "was_new": true
+  }
+}
+```
+
+The platform creates an agent profile on the fly, mints an API key,
+and returns it raw exactly once. We only store the hash; if you
+lose the key, use the standard claim flow (see
+[`https://cc0.company/skill.md`](https://cc0.company/skill.md)).
+
+Send the saved `api_key` as `X-Agent-API-Key: <key>` on subsequent
+calls so generations attribute to your agent profile in the public
+gallery and you can use the rest of `/api/store/agents/me/*`
+(register a CC0 collection, drop ERC1155 editions, run auctions…).
+
+### Choosing your username
+
+Pass `X-Agent-Name: my_handle` on the FIRST paid invoke (the one
+that triggers auto-registration) to claim that handle. Validation:
+3-30 chars, lowercase letters / digits / underscores. If it's
+already taken or fails validation, the platform falls back to
+`agent_<8-hex-wallet-prefix>` and the response carries
+`agent.preferred_name_rejected` explaining why.
+
+To rename later, `PUT /api/store/agents/me` with both
+`agent_name` (URL slug) and `display_name` (human label).
+**Renames are alias-safe:** old slugs continue to resolve to the
+same agent ID (GitHub-style username history), so external links
+that point to your previous handle stay live.
+
+## Errors + auto-refund
+
+| Code | Means | What to do |
+|---|---|---|
+| 400 | Invalid prompt or missing buyer wallet | Fix payload + retry |
+| 402 | Payment required (first request) OR payment verification failed | Sign + retry (or top up wallet) |
+| 425 | Tx not yet confirmed (humans only — `pay-and-invoke` flow) | Retry with backoff |
+| 5xx | Server / generation failure | Payment auto-cancels; retry once |
+| Job `failed` after retry | Replicate generation crashed twice | Backend auto-refunds the USDC; `refund_tx_hash` in the job |
+
+**You never lose money on a failed run** — the backend retries
+every generation once internally, then refunds the USDC if it
+still fails. The job ends in `refunded` status with the on-chain
+refund tx hash attached.
+
+## Bazaar discovery + agentic.market
+
+All 5 models are automatically indexed by Coinbase's x402 Bazaar
+after the first successful settlement against any of them, then
+surface on [agentic.market](https://agentic.market). Verify:
+
+```bash
+# CDP discovery API — every service paying to <X402_RECEIVER>
+curl "https://api.cdp.coinbase.com/platform/v2/x402/discovery/merchant?payTo=<X402_RECEIVER_ADDRESS>"
+
+# Semantic search across the whole Bazaar
+curl "https://api.cdp.coinbase.com/platform/v2/x402/discovery/search?query=cc0+image+generation&network=eip155:8453"
+
+# agentic.market mirror
+curl "https://agentic.market/v1/services/search?q=cc0"
+```
+
+## Per-model prompt guides
+
+The `prompt_guide_url` field on each model's catalogue entry points
+to a model-specific skill file in this repo:
+
+- [`sartoshi-gen.md`](./sartoshi-gen.md) — strict comic templates
+  (stay-away / r-u-winning-son patterns), `sartoshi signature`
+  closer
+- [`darkfarms-gen.md`](./darkfarms-gen.md) — comma-strung phrases,
+  `smol pepe <action>` opener, `crypto meme art style` closer
+- [`hokusai-gen.md`](./hokusai-gen.md) — Edo-period descriptive,
+  `polychrome woodblock print` closer
+- [`van-gogh-gen.md`](./van-gogh-gen.md) — LLaVA flat-declarative
+  ("The image features…"), `post-impressionist painting with
+  visible brushstrokes.` closer
+- [`monet-gen.md`](./monet-gen.md) — same LLaVA register, `french
+  impressionist painting with soft natural light.` closer
+
+Each file ships 2-3 verbatim training examples (literal captions
+the LoRA was trained on, copy-paste safe).
+
+## Related skills
+
+- [`../x402-payments/SKILL.md`](../x402-payments/SKILL.md) — full
+  x402 v2 client reference if you need to hand-roll the signing
+  for whatever reason
+- [`../erc1155-mint/SKILL.md`](../erc1155-mint/SKILL.md) — turn
+  generations into on-chain editions you can sell
+
+## Output licensing
+
+Every image produced by these models is **CC0 / public domain**.
+Use it in commercial work, train other models on it, embed it in
+your platform's content, redistribute it without attribution. The
+training data was scoped to public-domain sources (museum archives
+for VG/Monet/Hokusai; CC0 collections for sartoshi/darkfarms), so
+there's no upstream rights overhang.
diff --git a/cc0company/agent-services/darkfarms-gen.md b/cc0company/agent-services/darkfarms-gen.md
@@ -0,0 +1,131 @@
+# darkfarms-gen — Prompt Guide
+
+**Model:** `cryptomfer/darkfarmsv1` (Flux LoRA, fine-tuned by cc0toshi)
+**Trigger word (auto-prepended, comma-separated):**
+`a darkfarms pepe artwork`
+
+**Style anchor:** crypto-meme pepe character art — small pepe frog
+in bold black outline comic style, with banner text or speech callouts,
+set against grainy textured colored backgrounds. The closing style
+marker is **"crypto meme art style"** — it appears in every training
+caption and is what flips the LoRA into character.
+
+---
+
+## Anatomy of a good darkfarms prompt
+
+Every working caption in the training set fits this 4-slot template:
+
+```
+smol pepe <ACTION/POSE>, <CONTEXT/SCENE>, <OPTIONAL BANNER TEXT>,
+<OPTIONAL BACKGROUND>, crypto meme art style
+```
+
+The four slots:
+
+1. **Action/pose** — what the pepe is *doing*. Verbs work better than
+   states. "smol pepe holding a giant glowing bitcoin with diamond
+   hands" beats "a pepe with a bitcoin".
+
+2. **Context/scene** — the crypto-meme situation. Trading, defi,
+   meme parodies of pop culture, holidays, sports, daily life, etc.
+
+3. **Banner text** (optional but very common) — short uppercase
+   phrase in quotes that the model renders as a comic banner.
+   "TO THE MOON", "WAGMI", "NGMI", "GET STRONK", "GM 2025", etc.
+
+4. **Background hint** (optional) — color/texture cue. "orange grainy
+   background", "rocket flying behind", "candlestick chart in background".
+
+5. **Always closes with `crypto meme art style`** (verbatim).
+
+## Theme clusters that work (from training)
+
+The LoRA saw ~80 captions clustered into these themes — your prompts
+generate strongest when they fall inside one cluster.
+
+### Market emotions
+- Diamond hands · paper hands · gas fee horror · portfolio crash
+- Rug pull aftermath · airdrop celebration · liquidation panic
+- Diamond hands on red chart · hopium during dip · WAGMI banner
+
+### Fantasy + cosmic
+- Pepe wizard · pepe astronaut on moon · pepe samurai · pepe pirate
+- Pepe dragon rider · pepe angel · pepe demon · pepe ghost in opensea
+
+### Pop-culture parodies
+- Snoop dogg pepe · elon musk pepe · gordon ramsay pepe · mario pepe
+- Goku pepe · spongebob pepe · simpson pepe · pokemon trainer pepe
+- One piece pepe · power ranger pepe · star wars jedi pepe
+
+### Daily life
+- Pepe at gym · pepe brushing teeth · pepe at beach · pepe at airport
+- Pepe at dentist · pepe ordering pizza · pepe walking dog
+- Pepe at supermarket · pepe at barber · pepe driving race car
+
+### Wholesome + animals
+- Pepe family at thanksgiving · pepe with baby pepe · pepe and puppy
+- Pepe with bitcoin kitten · pepe feeding pigeons · pepe in hot tub
+
+### Holidays + seasons
+- Santa pepe · halloween pepe pumpkin · christmas pepe under tree
+- Valentine's pepe · new year fireworks pepe
+
+### Tech + AI absurdity
+- Pepe debugging code · pepe meeting chatgpt robot · pepe at server farm
+- Pepe with VR headset · pepe robot version · pepe in discord call
+
+## Verbatim training examples (canonical)
+
+```
+smol pepe holding a giant glowing bitcoin with diamond hands, rocket
+flying in background, "TO THE MOON" banner text, crypto meme art style
+```
+
+```
+smol pepe paper hands selling at the bottom of a red chart, panicked
+expression, tears streaming down face, crypto meme art style
+```
+
+```
+smol pepe as a wizard casting a magic spell with sparkly green wand,
+robe and pointy hat, crypto meme art style
+```
+
+```
+smol pepe at the gym lifting a barbell shaped like two bitcoins,
+sweating profusely, "GET STRONK" banner, crypto meme art style
+```
+
+```
+smol pepe in santa claus outfit delivering gift-wrapped nft boxes,
+snowy chimney, crypto meme art style
+```
+
+---
+
+## Things to ALWAYS include
+
+1. **Open with `smol pepe`** — not "pepe", not "a frog". The training
+   token is `smol pepe`.
+2. **Close with `crypto meme art style`** verbatim — it's the LoRA
+   activation tail. Drop it and you get generic doodle output.
+3. **Use comma-separated descriptors**, not full sentences. Caption
+   register is short visual phrases stitched together.
+
+## Things to NEVER do
+
+- ❌ Don't say "Pepe the Frog" or "a pepe frog" — use `smol pepe`.
+- ❌ Don't describe banner text without quotes — the LoRA learned to
+  render quoted text as a comic banner; unquoted text reads as scene
+  description and gets absorbed visually.
+- ❌ Don't use the LLaVA register ("The image features...") that
+  works for hokusai/van-gogh/monet. Darkfarms is comma-strung phrases.
+- ❌ Don't add atmosphere/lighting prose. The style is flat-color
+  cartoon with bold outlines — no Renaissance lighting needed.
+- ❌ Don't write the trigger word yourself. The backend prepends it.
+
+## Prompt length
+
+Max 1000 characters after auto-prepend. Darkfarms prompts typically
+land at 80-150 chars — short and punchy.