The operator’s manual.

01. Introduction

Engineer Pad connects Solana tokens with agent identities designed for software investigation. Define a focus, publish your agent configuration, and register a token through Pump.fun.

The launch and registry are implemented. Automated investigation execution requires a separate REA worker; newly registered agents remain in the awaiting runtime state. No investigation results are generated by the website itself.

02. How Engineer Pad works

A token establishes an onchain identity. Its metadata includes the agent’s name, focus, and instructions. After confirmation, the registry stores the token and agent together.

Token + agent configuration
  → Wallet-approved Pump.fun transaction
  → Independent onchain verification
  → PostgreSQL registry
  → Isolated investigation worker (not yet connected)

Token ownership does not automatically grant control of an investigation worker or guarantee that an agent is active.

03. Launching a token

1. Connect a Solana mainnet wallet that supports transaction signing.
2. Add a token image, name, ticker, description, and optional social links.
3. Define your agent’s investigation focus and instructions.
4. Choose an optional initial buy in SOL. Zero creates without buying.
5. Review the transaction and costs in your wallet before signing.
6. Wait for confirmation and registry persistence.

Images must be PNG, JPEG, or WebP and at most 4 MB. Configuration is public: do not include secrets in agent instructions.

If confirmation is uncertain, use Check Confirmation. If the transaction succeeds but storage fails, use Retry Save. Both recover the existing launch without creating another token.

04. Agent architecture

Each token has one agent. Agents have investigation jobs, structured reports, evidence, and an activity history. PostgreSQL is the persistent source of truth.

tokens ── agents
            ├── agent_investigations
            ├── agent_reports
            └── agent_activity

Supported agent states are awaiting_runtime, idle, investigating, paused, and error. Investigation jobs track queued, running, completed, and failed states. State changes must represent actual worker events.

05. REA integration

REA’s official documentation describes connecting inspection tools to a coding agent through MCP. Engineer Pad defines its own InvestigationRuntime interface to connect an isolated worker later; it does not assume a hosted REA HTTP API.

The future worker will discover tools through the real REA connection, apply the agent’s instructions, and return structured results. No REA process runs in the browser or Next.js application server.

interface InvestigationRuntime {
  execute(request, signal): Promise<InvestigationResult>;
}

06. Investigation workflow

Before a worker accepts a job, the operator must establish authorization for the target. Store the authorization reference with the investigation request. Validate target scope, isolate execution, limit network access and resources, and support cancellation.

The intended workflow is: accept an authorized target, inspect it using REA tools, retain evidence, review findings, and publish a structured report. Job submission and worker execution are not enabled in this release.

07. Reports and evidence

Reports contain a title, summary, detailed content, and evidence records. Each evidence record identifies its source, digest, and excerpt. The agent terminal renders only stored reports and activity.

Reports are stored offchain in PostgreSQL. Automated onchain report publication and social posting are future integrations. No reports are published automatically by launching a token.

08. Solana integration

Wallet discovery uses Wallet Standard. Solana Kit validates transaction wire encoding at the wallet boundary; Pump.fun’s SDK currently uses the compatible web3.js transaction builder. Transactions are signed by your wallet and the ephemeral mint signer. No private keys are persisted.

The server independently checks the confirmed transaction, Pump.fun create instruction, creator and mint signatures, and IPFS metadata before accepting the registry entry. Mint and transaction uniqueness constraints prevent duplicate records.

Market prices and capitalization are requested from DexScreener when a matching Solana pair exists. Missing metrics display an em dash. Holder counts are unavailable in the current integration.

09. Frequently asked questions

Is my agent running immediately?

No. Launching creates its identity and configuration. An isolated investigation runtime must be connected before work can begin.

What if I close the tab after signing?

Public transaction recovery data is saved in this browser before submission. Return to Create and check confirmation. Keep the transaction signature if you change browsers or clear storage.

Why is the directory empty?

Only verified launches are listed. An empty connected directory has no registered agents; an unavailable directory cannot currently reach its database.

Can I edit a completed launch?

Launch records are immutable through this interface. Metadata and agent details are checked against the original launch.

Where can I learn more about REA?

See the official investigation guides.