Submit
Bring your prompt and choose a Solana task. Define what a good answer should do.
THE SOLANA AGENT WORKSHOP
Good agents start
with better
instructions.
Give your agent a clearer brief. Choose a specialist, refine your instructions, and take a better starting point into your next build.
Built for Solana workflows / Powered by measurable progress
FORGE ENGINE / 001 PROMPT WORKSHOP
01 / INPUTYour instructionsbaseline.prompt
02 / CHALLENGEPut it to the testFive held-out challenges.
03 / GENERATIONKeep what works ↗Inspectable. Downloadable.
++INPUT → CANDIDATE → EVALUATIONANSWERS + CHECKS + AI ASSESSMENT
BUILD WITH INTENT↗TEST THE INSTRUCTIONS↗INSPECT THE EVIDENCE↗KEEP THE BETTER VERSION
01 / THE MECHANISM
Improvement has to
earn its place.
Choose a Solana specialist, refine your prompt, and inspect every change. Your generations stay ready to copy, download, and build on.
Forge
A specialist proposes clearer instructions, stronger boundaries, and an explicit output format.
Challenge
Run both prompts on the same held-out tasks. Compare exact detail checks and the AI judge’s assessment.
Keep
Save a generation with its answers, scores, model settings, and receipt. Download the prompt and inspect the evidence.
02 / THE ARENA
Bring your first prompt.
↗
Checking the evaluation service…
Meet Pip. Your prompt workshop companion.
A little curiosity. A clearer set of instructions.
EVALUATION RECEIPTAWAITING INPUT
⌘
Give your agent
a better brief.
Your baseline, candidate, actual answers, and evaluation report will appear here.
BASELINE→CANDIDATE
AI QUALITY ASSESSMENT
Original quality judgment
Candidate quality judgment
Candidate instructions
Objective scores check specific details and wording. Quality scores are subjective model judgments from one run, not independent verification.
03 / THE SPECIALISTS
A focused crew.
A clearer brief.
Three specialists. Three Solana workflows. Choose the right starting point for your next generation.
04 / YOUR GENERATIONS
A record you can inspect.
Reports are private to your signed-in wallet and retained for 30 days. Download a receipt to keep it longer.
05 / SOLANA ACCESS
Built around
$FORGE.
Connect a Solana wallet. Sign in with a message, check your allowance, and put your agent’s instructions to work.
Before the token launches, wallet sign-in provides a small evaluation allowance. Holder requirements will appear here when configured.
$FORGE
SOLANA / ACCESS TOKEN- Token mint
- Token not launched
- Holder threshold
- Prelaunch wallet access
- Evaluation allowance
- Loading…
- Access status
- Wallet sign-in required
Connecting and signing in do not authorize transactions or token transfers. Eligibility is checked by the service.
THE DETAILS
Before you forge.
What does Agent Forge actually improve?
Your agent’s instructions. An optimizer proposes a candidate using the task brief and training examples. Both versions then answer the same five held-out cases. Inspect those answers before deciding what to use.
What happens on Solana?
Your Solana wallet signs a login message. When holder access is enabled, the server checks the official token’s balance before allowing a job. AI processing happens offchain. No trading or token transfer is requested.
What do the checks measure?
Each case has literal-detail and uncertainty checks. A blinded AI judge also scores answer quality separately. A candidate is accepted only when its average quality score improves and no case loses objective checks. The judge is not an independent verifier, and results can vary on replay.
Can I use the result in my existing agent?
Yes. Copy the revised instructions or download a JSON receipt. Test the instructions with your chosen model before deploying them.
Where are my prompts stored?
Submitted prompts and examples are sent to Cloudflare Workers AI for processing. Cloudflare’s processing policies apply. Reports are stored privately for your wallet for up to 30 days. Wallet addresses and usage counters are retained to enforce allowances. Reports are not posted publicly or written onchain.