Core concepts
Agents
Nine agents, six implemented. What each one collects, and what it needs to be able to run.
An agent owns one kind of question. A single analysis usually runs several: a token analysis runs the market agent and the risk agent, and the risk agent scores what the others found.
| Agent | What it collects | Status |
|---|---|---|
market | Token identity, supply, holders, price and market data | Implemented |
contract | Verification, source availability, owner powers | Implemented |
risk | Scores everything the others found, into signals and a level | Implemented |
portfolio | Screens each token a wallet holds, using the same rules | Implemented |
project | Chain-level statistics and network health | Implemented |
site | Domain age, certificates, archives, page claims | Implemented |
launch | Registered, not implemented — it cannot be executed | Planned |
builder | Registered, not implemented | Planned |
community | Registered, not implemented | Planned |
One set of scoring rules
The risk agent and the portfolio agent call the same pure functions over the same facts. A token screened inside a wallet is judged exactly as one analysed on its own — there is no second copy of a rule written for a new surface.
Agent status
status is derived from the live health of the services an agent depends on, not from stored metrics:
- active
- Implemented, and every dependency is answering.
- degraded
- Implemented, but a dependency is not live. It will produce fewer findings.
- not_implemented
- Registered in the runtime with no implementation. It cannot run.