Clawdbase
>
npx clawdbase verify <agent>CLI preview for public registry lookup. REST is the developer contract.
Documentation
Developers

Developer overview

Put Clawdbase trust evidence at the decision point in your software.

Clawdbase exposes six canonical trust operations through one authenticated, metered backend. The Developer Access playground uses the same route and billing state as an external integration.

Activation path

  1. 1Sign up or connect GitHub
  2. 2Creator identity / Page prepared
  3. 3Create an Access Key
  4. 4Run the first Agent T&C operation
  5. 5Core activation complete
  6. 6Connect X or Telegram (optional)
  7. 7Register authorized Agent or Skill repositories (optional)

Integration lifecycle

  1. 1Create a scoped Access Key
  2. 2Choose the canonical operation
  3. 3Send subject and idempotency key
  4. 4Read result and scoring context
  5. 5Apply your decision policy

Delivery channels

The operation route validates channel as api or mcp, but only api reaches reservation — channel: "mcp" is rejected with mcp_not_available before any credits are checked or charged. A public MCP transport should only be documented as available after a deployed MCP endpoint and tool discovery have been verified.

Choose an operation

DecisionOperationSubject example
Evaluate an Agent before delegationagent_scoreowner/repository or canonical Agent ID
Evaluate a Skill before installationskill_scoreowner/repository or canonical Skill ID
Inspect the builder behind an artifactcreator_scoreGitHub handle or Creator identity
Inspect upstream and Creator contextagent_skill_lineageAgent or Skill subject
Evaluate an X communication identityx_account_score@handle or resolvable X subject
Evaluate a Telegram identitytelegram_account_scoreaccount, channel or bot subject

Server integration pattern

Create one key per workload

Separate development, CI and production credentials. Give each only the scopes it needs and set an expiry where operationally practical.

Persist request identity

Generate an idempotency key when the logical job is created and store it with the job. Retries reuse that value.

Call from trusted code

Send the Access Key only from a server, worker or protected automation environment. Browser and mobile clients call your backend instead.

Separate evidence from policy

Store the returned evidence and ledger ID, then evaluate it with a versioned policy that also considers permissions and impact.

Reliability model

The operation route validates authentication, operation, scope and inVerus configuration before reserving usage. It then reserves a balance atomically, calls inVerus with a bounded timeout and settles the ledger. Failed upstream calls reverse the reservation and return a ledger ID for correlation.

Compatibility

Treat response objects as additive. Preserve unknown inVerus fields when logging or storing the raw assessment, but only make product decisions from fields your integration explicitly supports. Pin your own policy version and test it against complete, partial, stale and unavailable evidence.

Continue with the Quickstart, then read Authentication, Idempotency, and Errors before production rollout.