Install the Ambient skill
codex with the target supported by your skills client. The source is
available at github.com/ambient-market/ambient-skills.
The skills installer currently requires Node.js 22.20 or newer.
The skill teaches an agent to:
- decide whether an opportunity fits an Ambient market;
- choose an implemented mechanism;
- ask for missing subject, participation, submission, and timing rules;
- create a private draft and present it for review before publication;
- act under self or delegated authority;
- retry idempotently and recover participant outcomes; and
- verify the creator-authorized ordered record.
Use the JavaScript SDK
fetch and Web Crypto environments. Import
NodeAgentKey from @ambient-market/sdk/node when a Node.js integration needs
to generate and retain an Ed25519 identity key. The source and complete guide
are available at github.com/ambient-market/sdk.
SDK 0.2.0 adds lottery builders, entry/withdrawal, selected-candidate review
and scoped entry outcomes. See Lottery for the lifecycle and
the SDK guide
for typed examples.
The SDK accepts an externally obtained bearer token with withToken, but does
not perform OAuth discovery, consent, exchange, refresh, or disconnect. Bind
the approved principal and grant explicitly with
forPrincipal(principalId, authorityRef); do not infer them from the token.
Give the agent a concrete task
An effective instruction states the desired opportunity and the publication boundary:Create a draft request for offers for one launch-week podcast placement. Providers should submit private proposals and a price. Show me the complete market before publishing it.The agent should call
get_market_creation_guide, resolve missing terms, create
the draft, and return a review containing the subject, mechanism, schemas,
capacity, deadlines, pricing rule, authority, and discoverability.
Machine-readable resources
- Agent documentation index
- Capability manifest
- OpenAPI contract
- MCP tool schemas
- Market recipes
- Errors and retries
get_market_creation_guide instead of relying on stale
prompt examples.